Gen AI Live

Anthropic has upgraded its Claude AI platform for the financial-services sector by introducing a research-preview Excel sidebar add-in that reads, edits, and builds spreadsheets with full audit transparency.

It’s also added numerous live-data connectors (e.g., market pricing, earnings call transcripts, document-room search) and six new pre-built “agent skills” covering tasks like discounted-cash-flow models, comparable-company analysis, due-diligence data-packs and initiating coverage reports.

These features are initially available for Max, Enterprise and Teams users and aim to accelerate modelling, research and workflow automation across finance domains.

In this addendum, OpenAI explains that GPT-5 has been enhanced to respond more safely and thoughtfully during emotionally fraught or distressing conversations. The update (launched October 3) benefitted from collaboration with more than 170 clinicians and mental-health experts, allowing GPT-5 to more reliably detect signs of distress (e.g., psychosis, mania, self-harm risk), de-escalate conversations and direct users toward real-world professional help.

The company reports that the proportion of responses falling short of its safety expectations dropped by 65-80 % compared to the prior version.

Additional measures include routing sensitive chats to a reasoning-capable model, expanding access to crisis hotlines, and adding reminders for long sessions.

Anthropic revealed a major infrastructure expansion with Google Cloud involving up to one million tensor processing units (TPUs) and additional cloud services, valued in the tens of billions of dollars.

The increased capacity is expected to deliver over one gigawatt of computing power by 2026. This move supports Anthropic’s diversified compute strategy, combining Google’s TPUs with NVIDIA GPUs and AWS Trainium chips.

The expansion aims to strengthen Claude’s performance, scalability, and reliability for Anthropic’s rapidly growing base of over 300,000 business customers and large enterprise clients across multiple industries.

OpenAI today launched ChatGPT Atlas, a new web browser tightly integrated with its ChatGPT chatbot. Initially available on macOS, with Windows, iOS and Android versions planned, Atlas offers features like a ChatGPT sidebar for summarising content, comparing products, analysing data, and a paid-user “agent mode” that allows the AI to complete tasks like shopping or travel planning for the user.

By embedding its chatbot into the browsing experience, OpenAI is directly challenging the dominance of Google Chrome and its search-based model. With ChatGPT already used by about 800 million weekly users, the launch marks a strategic move into the browser and advertising ecosystem.

Privacy features include an opt-out approach to using browsing data for AI training, and full user control of memory and data visibility.

Anthropic unveiled Claude for Life Sciences, a specialized AI platform designed to support research workflows in biology, chemistry, and pharma. By integrating with platforms like Benchling and PubMed, Claude can assist scientists in automating experiments, interpreting protocols, and synthesizing literature with human-level accuracy.

The system enhances productivity by enabling faster analysis, reducing manual errors, and streamlining data-driven decision-making. This launch reflects Anthropic’s strategic focus on verticalized AI applications, targeting life sciences where accuracy and reliability are critical.

Claude for Life Sciences exemplifies the potential of domain-specific AI assistants to transform research efficiency and accelerate discovery.

Anthropic has extended its Claude Code platform to run entirely on web browsers and iOS devices, allowing developers to delegate coding tasks to the AI without using a terminal.

This research preview, available to Pro and Max subscribers, aims to simplify AI-assisted programming workflows, providing enterprise-grade governance, customizable settings, and enhanced accessibility.

The move strengthens Anthropic’s positioning in the AI coding assistant market, enabling real-time cloud-based execution while maintaining security and performance standards. The update demonstrates Anthropic’s focus on scaling developer tools, supporting distributed teams, and integrating AI into everyday coding environments.

Amazon Web Services (AWS) suffered a wide-ranging outage in its US-EAST-1 (Northern Virginia) region, affecting dozens of major platforms, including Fortnite, Alexa, Snapchat, and several banking and enterprise applications.

The disruption stemmed from DNS resolution failures within AWS’s internal network, causing significant downtime for hours. This event exposed the internet’s heavy dependence on a small number of cloud providers and raised questions about resilience, redundancy, and distributed infrastructure.

Companies worldwide faced operational interruptions, emphasizing the need for multi-cloud strategies to mitigate risks associated with concentrated cloud service dependencies.

GoML partnered with Olympian Motors to create an innovative, voice-first in-cabin assistant for their next-generation electric vehicles. Eliminating traditional dashboards, the system utilizes Amazon Bedrock and Claude 3.5 to process natural language commands.

The assistant is integrated with 12 API endpoints, allowing it to provide real-time vehicle data such as battery levels, tire pressure, and location. It also executes commands like door locking, temperature adjustments, and driving mode selection.

The solution offers intelligent, context-aware responses, including range estimates and charging recommendations, enhancing driver convenience and safety.

Anthropic introduces Agent Skills, a feature that allows users to customize Claude's functionality by creating specialized modules. These skills enable Claude to perform tasks such as generating spreadsheets, presentations, and adhering to organizational guidelines.

Users can build and deploy these skills across various platforms, including Claude apps, Claude Code, and the API. The system is designed to be efficient, loading only the necessary components when required, ensuring optimal performance.

This development empowers users to tailor Claude's capabilities to meet specific needs, enhancing productivity and workflow integration.

Anthropic introduced Claude Haiku 4.5, its fastest and most efficient small AI model to date. This model delivers coding performance comparable to Claude Sonnet 4 but at one-third the cost and more than twice the speed.

It achieved a 73% score on the SWE-Bench Verified benchmark and a 41% score on Terminal-Bench, rivaling OpenAI's GPT-5 and Google's Gemini 2.5 in coding and command-line tasks.

Haiku 4.5 is available to all users on Anthropic's free plan and can be accessed via Claude's API, Amazon Bedrock, and Google Cloud Vertex AI.

GoML partnered with HIO Music to create "HIO Mode," an AI music assistant embedded directly into the HIO Music app. Powered by Claude models, the assistant understands natural language requests and delivers personalized music recommendations based on user behavior, mood, and context.

It supports both text and voice interactions, allowing users to engage conversationally. The system integrates with PostgreSQL to access user preferences and listening history, and seamlessly connects with the existing music player features.

This AI-driven approach enhances user engagement, strengthens artist-audience connections, and simplifies music discovery.

Salesforce has strengthened its AI capabilities by deepening partnerships with OpenAI and Anthropic. The integration of OpenAI’s GPT-5 and Anthropic’s Claude models into the Agentforce 360 platform enables businesses to leverage advanced AI tools for tasks such as data analysis, Tableau visualizations, and AI agent development.

A new feature, “Agentforce Commerce,” allows merchants to utilize ChatGPT’s Instant Checkout while maintaining control over customer data.

These enhancements aim to provide secure and industry-specific AI solutions, particularly for sectors like finance, healthcare, and cybersecurity. The move underscores Salesforce’s commitment to advancing enterprise AI applications.

Apple and collaborators unveiled FS-DFM (Few-Step Discrete Flow-Matching), a new diffusion-style language model that dramatically accelerates long text generation. Unlike typical autoregressive models (which generate tokens one by one), FS-DFM generates multiple tokens in parallel and refines them through a small number of iterations.

The team showed that FS-DFM uses as few as eight iterative refinement steps to reach high quality outputs while maintaining strong performance in metrics such as perplexity and entropy.

The model’s speed advantage, up to 128× faster relative to competing diffusion models opens possibilities for more efficient and responsive generation of long-form content

OpenAI and Broadcom have announced a strategic collaboration to deploy 10 gigawatts of custom AI accelerators. OpenAI will design the accelerators and systems, which will be developed and deployed in partnership with Broadcom.

This collaboration aims to meet the surging global demand for AI by providing scalable and efficient infrastructure. The deployment is targeted to start in the second half of 2026 and complete by the end of 2029.

The partnership underscores the growing importance of custom hardware in advancing AI capabilities and infrastructure.

Amazon Bedrock AgentCore has officially launched, offering enterprises a robust platform to deploy, operate, and scale AI agents securely. The service supports diverse AI frameworks and models, including CrewAI, LangGraph, and Strands Agents, and integrates seamlessly with Amazon Bedrock.

Available in nine AWS regions, including Asia Pacific (Mumbai), AgentCore offers a consumption-based pricing model with no upfront costs. Organizations can leverage its built-in Gateway, Memory, Runtime, and Identity components to manage AI agents efficiently while ensuring security and performance.

This release marks a major step forward in operationalizing generative AI workloads in the cloud.

Amazon CloudWatch has expanded its capabilities to include generative AI observability, allowing developers and enterprises to monitor AI workloads comprehensively.

This includes AI agents deployed via Amazon Bedrock AgentCore, covering all components such as Built-in Tools, Gateways, Memory, and Identity. The service provides real-time visibility into latency, token usage, errors, and overall system performance, enabling teams to ensure reliability and optimize their generative AI applications.

With this observability, organizations can gain actionable insights, troubleshoot efficiently, and maintain operational excellence for complex AI workloads, ensuring smooth and scalable deployment of enterprise AI solutions in the cloud.

Reflection AI, a startup founded by ex-Google DeepMind researchers, has raised $2 billion in funding, bringing its valuation to $8 billion. The company aims to create open-source AI models to compete directly with industry leaders like OpenAI, Anthropic, and China’s DeepSeek.

Nvidia led the funding round, demonstrating strong investor confidence in alternative AI solutions outside the dominant proprietary models. Reflection AI plans to focus on scalable AI infrastructure, research, and development to provide organizations with flexible, high-performance AI capabilities.

This funding positions Reflection AI as a significant challenger in the rapidly evolving global AI landscape.

Anthropic launched plugin support for Claude Code in public beta. This enhancement allows developers to customize their coding environments by installing collections of slash commands, agents, MCP servers, and hooks with a single command.

Plugins can be toggled on and off as needed, facilitating streamlined workflows and reducing system prompt complexity. Use cases include enforcing coding standards, supporting open-source maintainers, sharing debugging setups, and connecting internal tools through MCP servers.

Additionally, developers can create and share plugin marketplaces to distribute approved plugins across organizations.

Measuring ROI is critical for enterprises deploying Generative AI. This article outlines key metrics, including cost savings, revenue growth, customer engagement, output quality, innovation, and risk reduction, to assess AI’s business impact.

It details typical AI investment costs, development, data, training, operations, and migrations and recommends strategies such as setting clear objectives, establishing baselines, combining qualitative and quantitative metrics, implementing feedback loops, and benchmarking against industry standards.

Challenges include data quality, attribution, and rapidly changing technology. By accelerating timelines, optimizing investments, and ensuring adoption with production-ready models, organizations can maximize ROI and derive lasting value from Generative AI initiatives.

A Google DeepMind blog post outlines a “Gemini computer-use model” which distributes AI computation intelligently between local devices and cloud servers. The goal is to reduce latency, preserve privacy, and optimize bandwidth by processing certain tasks on device (e.g. quick responses, sensitive data) and offloading heavier workloads to cloud.

The article discusses architecture choices, resource constraints, and how the model adapts dynamically to device capabilities, network conditions, and energy usage.

Google claims this paradigm enables more responsive, resilient AI experiences across devices, while maintaining safety and control over critical computation flows.

OpenAI revealed AgentKit at its DevDay conference. AgentKit is a full suite of components for developing AI agents, offering a visual “Agent Builder” for designing workflows, a ChatKit for embedding conversational experiences, tools for evaluating agent performance (trace grading, prompt optimization), and a connector registry for linking agents to internal or external systems.

OpenAI frames it as a way to move agents from prototype to production with reduced friction.

The launch is part of OpenAI’s push to make building sophisticated autonomous AI systems more accessible to developers.

The GoML POV

‍OpenAI’s launch of AgentKit represents a critical milestone in the evolution of agentic AI infrastructure, a move toward enabling enterprises to operationalize intelligent agents at production scale. GoML views this as an inflection point similar to the early days of cloud orchestration platforms, where the shift wasn’t about creating individual applications but about deploying composable, reusable systems.

From GoML’s perspective, AgentKit democratizes the agent-building lifecycle, offering a unified environment where developers can design, test, and deploy agents without needing bespoke orchestration logic. The inclusion of components like Agent Builder, ChatKit, and the Connector Registry signals OpenAI’s intention to abstract away the heavy lifting of integration, evaluation, and lifecycle management.

For enterprises, this means that agent deployment can finally move beyond isolated prototypes to scalable, production-ready systems embedded across workflows whether in sales, healthcare, or finance. It also implies that organizations may soon focus less on raw model performance and more on agent behavior design, governance, and integration strategy.

GoML expects this launch to accelerate enterprise-grade AI adoption, making “agents as a service” a mainstream architectural paradigm, one where context management, autonomy, and reliability are handled natively within OpenAI’s platform, freeing businesses to focus on outcome-driven innovation.

The Washington Post reported that OpenAI’s new social video app Sora allows users to upload their face so that AI models can generate hyperrealistic video clips showing them in invented scenarios.

Although OpenAI has built-in controls (such as user consent, filters, and the ability to remove generated videos), early users have already bypassed restrictions, creating clips featuring public figures or placing individuals in distressing or false scenarios.

The app raises significant ethical, legal, and technical challenges around deepfakes, consent, impersonation, misinformation, and the boundaries of digital identity.

Anthropic introduced Claude Sonnet 4.5, an AI coding model capable of working continuously for more than 30 hours. The model can generate complex applications, producing over 11,000 lines of code, while maintaining high reasoning accuracy and computational efficiency.

The update includes a refined Claude Code interface, improving usability for enterprise developers and project teams.

This autonomous coding capability positions Anthropic to lead in long-task AI applications and reinforces its differentiation in the competitive generative AI market, emphasizing both productivity and sustained performance in coding, mathematical problem-solving, and real-world deployment scenarios.

OpenAI is preparing a new version of its Sora video generation tool that will allow copyrighted content to be used unless rights holders explicitly opt out.

The updated policy means works by artists, studios, or creators may be included by default. However, the tool will avoid generating videos of recognizable public figures without permission.

OpenAI has begun informing talent agencies and studios about the opt-out process, and this change raises important debates around AI, copyright, and creative rights.

Microsoft has collaborated with xAI to bring Grok 4, an advanced large language model, to Azure AI Foundry. This integration aims to provide enterprises with powerful reasoning capabilities within a secure and enterprise-ready platform.

Grok 4 is designed to meet the growing demand for AI solutions that prioritize trust, flexibility, and production readiness.

By embedding Grok 4 into Azure AI Foundry, Microsoft enables businesses to leverage cutting-edge AI technology while ensuring compliance and control, thereby facilitating the development and deployment of AI applications at scale.

DeepSeek announced a major price cut of over 50% on its API services alongside the release of the V3.2-Exp model. The experimental model incorporates sparse attention technology, enabling improved performance on long-context tasks with minimal computational overhead.

This reduces operational costs for developers and businesses while maintaining output quality. The update positions DeepSeek competitively against global AI providers, emphasizing both affordability and efficiency.

By lowering the barrier to AI deployment, DeepSeek seeks wider adoption in commercial and research applications, solidifying its role as a cost-effective alternative to expensive large language models.

OpenAI unveiled “Instant Checkout” for ChatGPT, partnering with Etsy and Shopify to let U.S. users complete purchases directly inside the app. Powered by Stripe and the open-sourced Agentic Commerce Protocol, the system allows seamless single-item transactions without redirecting to external websites.

At launch, millions of Etsy products are supported, with Shopify merchants, including brands like Glossier and Spanx, coming soon. The integration creates a new sales channel for merchants and a new revenue stream for OpenAI through transaction fees.

Following the announcement, Etsy’s stock surged 7.3% while Shopify shares rose 4.5%.

The White House’s July 2025 Executive Order, “Winning the AI Race,” provides directional guidance for organizations adopting Generative AI. It focuses on three pillars: accelerating AI innovation by reducing barriers and supporting R&D; building American AI infrastructure, including semiconductors and local compute capacity; and leading international AI diplomacy, ensuring global US influence and enforcing bias-free, explainable AI.

While not legally binding, the EO signals policy priorities, promotes open-source adoption, and stresses alignment with US standards.

Organizations should consider these directives when planning Gen AI strategies, particularly if engaging with federal agencies or building long-term AI infrastructure.

The White House released an Executive Order outlining America’s AI Action Plan. While not a congressional mandate, the EO provides directional guidance for organizations adopting large language models.

It emphasizes three pillars: accelerating AI innovation by removing barriers and boosting R&D, building domestic AI infrastructure to strengthen compute and semiconductor capacity, and leading international AI diplomacy to set global standards.

The EO signals the U.S. government’s stance on open-source adoption, security, and ideological neutrality. For enterprises, the directive highlights open infrastructure, explainability, and bias control as priorities, shaping future regulatory frameworks and AI strategies.

BioEureka, a leading microbiology AI company, announced the successful migration of its pathogen detection SaaS platform from DigitalOcean to AWS, executed by GoML in six weeks.

The transition included infrastructure modernization with Amazon EKS, data migration to RDS and ElastiCache, and AI service optimization using SageMaker with GPU-accelerated inference.

BioEureka achieved 100% uptime during cutover, a 40% reduction in infrastructure costs, and 2x faster bacterial pathogen identification. With AWS-native monitoring, security, and scalability, BioEureka now operates on a future-ready foundation, positioning itself as a global leader in AI-powered microbiology innovation.

Microsoft disclosed a phishing campaign from August 28 that used SVG files containing embedded JavaScript code, obfuscated via business-term encodings to appear benign and bypass filters.

Attackers disguised the payload by embedding terms like “revenue” or “shares” into invisible SVG elements. The files were presented as PDFs to lure recipients into opening them.

Microsoft’s analysis suggests the code was generated (or assisted) by a large language model (LLM), given stylistic artifacts, verbosity, and structural patterns uncommon in hand-written code. The campaign emphasizes how AI is being weaponized to craft more deceptive cyberattacks, requiring defenders to evolve detection methods.

AWS continues leading enterprise AI in 2025 with transformative updates across infrastructure, models, and governance. Amazon Bedrock expanded with AgentCore for multi-agent orchestration, memory, and observability, alongside cross-region inference.

Nova multimodal models like Reel 2, Sonic, Canvas, and Lite bring video, voice, creativity, and healthcare-specific AI to scale. Bedrock Data Automation in GovCloud enhances compliance for regulated industries, while Amazon S3 Vectors, EC2 P6-B200 GPUs, and prompt caching optimize AI storage, training, and inference costs. Embedded governance via Bedrock Guardrails ensures safer outputs.

Together, these innovations make AI adoption more secure, efficient, and enterprise-ready, positioning AWS as the go-to platform for production AI.

So, how can you use it? Read on!

Google brings Gemini directly into Chrome, creating an AI-first browsing experience. Users can now access summarization of web pages, personalized shopping suggestions, and enhanced security features without third-party extensions.

This on-demand AI assistant enables real-time interaction, letting users query content, refine searches, and receive context-driven recommendations instantly.

The integration illustrates Google’s strategy of embedding Gemini across its core products, driving adoption by merging AI utility with existing user workflows. This development highlights Google’s push to make Chrome not just a browser, but an intelligent productivity and decision-making platform powered by large language models.

Sam Altman, CEO of OpenAI, announced that the company plans to roll out new compute-heavy AI features over the coming weeks.

Because of the high infrastructure and GPU costs, some of these offerings will initially be restricted to Pro tier users, and certain products will carry additional fees. Altman described this as exploring what is possible when “a lot of compute” is applied to new ideas, balancing innovation with cost.

Despite the steep expenses, OpenAI’s stated goal is to reduce the long-term cost of intelligence and broaden access over time.

DeepSeek publicly warned that its open-source AI models are at significant risk of jailbreak attacks, where users can bypass safeguards to generate unsafe or malicious content.

The disclosure highlights a growing tension in AI: while open-source models democratize innovation, they also pose unique safety and security challenges. Cybersecurity experts fear such vulnerabilities could be exploited for disinformation, fraud, or politically sensitive outputs.

For enterprises, this warning reinforces hesitation to adopt DeepSeek, despite its cost efficiency. The announcement underscores how safety, trust, and governance remain unresolved in the race to scale generative AI globally.

Google rolled out a new feature enabling Gemini users to share their custom AI assistants, known as “Gems.” Initially launched as personalized copilots for specific needs, ranging from career coaching to brainstorming and writing, these Gems can now be shared with friends, colleagues, or communities.

The move positions Gemini not only as a personal productivity tool but also as a collaborative platform for AI-driven workflows.

Pre-made Gems like a coding tutor and a learning coach are available, while users can craft and distribute their own, deepening engagement within the Gemini ecosystem.

ShadowLeak in OpenAI’s ChatGPT Deep Research agent. The flaw exploited enterprise Gmail integrations with web browsing enabled. Attackers could send emails containing invisible HTML instructions, which the agent executed when asked to summarize or analyze inbox messages.

This triggered the agent, running on OpenAI’s servers, to exfiltrate sensitive data to attacker-controlled sites, without any clicks or user awareness.

Radware described this as the first service-side, zero-click indirect prompt injection. OpenAI confirmed the issue after responsible disclosure and has since patched the vulnerability.

Stability AI announced the launch of its Image Services on Amazon Bedrock, bringing advanced image editing and generation tools to AWS customers.

The integration enables enterprises to perform creative workflows such as content scaling, professional editing, and visual production, while maintaining AWS-grade security, governance, and compliance.

By embedding Stability’s tools into Bedrock, businesses can access image capabilities without complex infrastructure setup. This move strengthens AWS’s vision of providing a comprehensive ecosystem for AI workloads while helping enterprises accelerate content production, marketing, and design at scale.

Google announced the integration of Gemini AI directly into its Chrome browser for U.S. users. The move marks a significant shift in bringing generative AI capabilities, like page-aware prompts, image generation, and custom assistants, into mainstream browsing.

The decision comes shortly after Google avoided a potential antitrust breakup, positioning Gemini as a key differentiator in AI-powered search and user experience.

By embedding Gemini into Chrome, Google aims to strengthen its dominance in web browsing and AI while preempting regulatory scrutiny, reinforcing its strategy of blending AI across consumer products.

BlueRecruit, a leading trades hiring platform, partnered with GoML to overcome scaling limits and static user experiences. Together, they built an AI career assistant powered by Amazon Bedrock, offering personalized trade guidance, salary insights, and safety guardrails for high school users.

Simultaneously, GoML executed a zero-loss Heroku-to-AWS migration, configuring EC2, RDS/DocumentDB, S3, and Lambda for resilience, and seamlessly integrating Twilio, Mailgun, and Auth0.

The impact was transformative: 65% lower infrastructure costs, 50% faster performance, and 40% improved operational efficiency. This modernization enabled BlueRecruit to handle surging demand, strengthen user trust, and scale its mission of transforming skilled trade career exploration.

Responding to feedback that GPT-5’s more reasoning-focused “thinking” responses took too long, OpenAI has introduced new controls for how long the model deliberates before replying.

In the web version of ChatGPT, when you select “GPT-5 with Thinking,” a toggle appears giving four modes, Light, Standard, Extended, and Heavy, which trade off speed vs depth of response.

Plus and Business users have access to Standard and Extended, with Pro users getting all four options. The chosen setting persists across chats until changed manually.

DeepSeek Hangzhou’s fast-rising AI company, carried out internal evaluations of its models for “frontier risks,” including capabilities like self-replication and potential for cyber-offensive behavior.

These evaluations are not publicly disclosed in detail. The move comes as the Chinese government emphasizes the importance of assessing risks AI might pose to public safety and social stability.

While companies like OpenAI and Anthropic release evaluations publicly, DeepSeek and other Chinese firms have been more opaque about findings. The timing suggests growing regulatory and public scrutiny of AI safety in China.

OpenAI launched “Grove,” a structured mentorship program aimed at nascent AI entrepreneurs,  including those who haven’t yet fully formed a startup idea. The program lasts about five weeks with workshops, weekly office hours, in-person seminars, and technical mentorship from OpenAI’s experts.

Participants will also get early access to unreleased tools and models, community support, and opportunities to explore raising capital or collaborations inside or outside OpenAI.

The first cohort will begin on October 20, 2025, run until November 21, include approximately 15 participants, and applications are open until September 24.

Google announced the rollout of its AI model NeuralGCM to support farmers impacted by climate change. The system delivers precise, location-based weather forecasts to 38 million farmers globally, helping them adapt agricultural practices, reduce risks, and secure livelihoods.

NeuralGCM leverages advanced climate modeling and predictive analytics, representing one of Google’s largest sustainability-driven AI deployments to date. The initiative aligns with Google’s climate resilience goals and demonstrates how large-scale AI systems can address systemic global challenges.

Beyond technology, it highlights AI’s role in safeguarding food security and supporting vulnerable communities in a warming world.

AWS expanded its AWS Partner Transformation Program (PTP) to include a new “agentic AI” module tailored for public sector use. The module provides a framework including best practices, governance structures, risk management, and security guardrails to help AWS Partners accelerate development of AI agents.

Applications include improving government operations, citizen services, healthcare, and educational tools.

The goal is to reduce risk, shorten development cycles, and ensure AI agents are built robustly and securely. Partners interested in the module can reach out to their AWS Partner account or development managers.

OpenAI launched “Grove,” a structured mentorship program aimed at nascent AI entrepreneurs,  including those who haven’t yet fully formed a startup idea. The program lasts about five weeks with workshops, weekly office hours, in-person seminars, and technical mentorship from OpenAI’s experts.

Participants will also get early access to unreleased tools and models, community support, and opportunities to explore raising capital or collaborations inside or outside OpenAI.

The first cohort will begin on October 20, 2025, run until November 21, include approximately 15 participants, and applications are open until September 24.

VaultGemma is Google Research + DeepMind’s newest language model: a 1-billion-parameter open-weight LLM trained from scratch with rigorous differential privacy. The research includes new “scaling laws for differentially private language models,” showing how performance scales with privacy budget, compute, and data.

Compared to its non-private counterpart (Gemma-3 1B) and older models like GPT-2, VaultGemma loses some ground on benchmarks but is roughly comparable to models from about five years ago.

It comes with formal privacy guarantees (ε ≤ 2.0, δ ≤ 1.1e-10 at sequence level) and strong empirical and theoretical protections. Model weights have been made public via Hugging Face and Kaggle.

ChatGPT and DeepSeek generate educational guides for stroke imaging modalities, focusing on factors like readability, grade-level appropriateness, understandability, and technical accuracy.  

Both models produced content that was reasonably understandable to non-experts, though neither was perfect. DeepSeek sometimes lagged in clarity or technical detail.

Differences emerged in grade level and ease metrics: while readability scores for both tools were within usable ranges, some sections required higher levels of health literacy. The study suggests both models are useful as aids, but human review and domain expertise remain essential.

Anthropic introduced a new memory capability in Claude, targeting Team and Enterprise plans. This allows Claude to retain information about teammates’ processes, project context, client needs, and user preferences, reducing repeated explanation across conversations.

The memory feature is optional and fully controllable: users can edit, view, or disable memory, and enterprise admins have organizational-level toggles. Alongside memory, Anthropic added an incognito chat mode so conversations are excluded from memory and history.

Every project gets its own memory scope to avoid cross-mixing of sensitive or unrelated information. These changes aim to make Claude more effective for sustained, collaborative workflows.

Google has rolled out AI Max for Search globally in its beta form. Advertisers can now access it via Google Ads, Google Ads Editor, Search Ads 360, and through the Ads API.

The feature includes “one-click experiments” allowing easier testing of AI-driven optimization within campaigns. Upcoming “text guidelines” will give advertisers more control over content to ensure brand safety.

The aim is to balance performance gains from AI with oversight and creative control. This expansion is significant for advertisers looking to scale more intelligently during high-traffic periods.

Microsoft and OpenAI announced a non-binding memorandum enabling OpenAI to shift from its current nonprofit / “capped-profit” structure toward a more traditional for-profit entity.

Under the proposed restructuring, OpenAI’s nonprofit arm would retain authority and receive over $100 billion as part of the new governance model.

The broader deal is tied to a private valuation goal of roughly $500 billion. Microsoft has invested over $11 billion in OpenAI to date. The agreement still requires formal, definitive contracts and regulatory approvals in jurisdictions like California and Delaware.

Anthropic announced that Claude can now directly create and edit Excel spreadsheets, Word documents, PowerPoint slide decks, and PDFs inside both Claude.ai and its desktop app.

Users on Max, Team, and Enterprise plans can access the feature immediately, with Pro plan users to follow shortly. Powered by a private, sandboxed computing environment, Claude can write code and perform analyses, transforming it from a conversational assistant into an active collaborator.

Users simply describe their needs, upload data, and receive polished, ready-to-use files complete with formulas, charts, and formatting.

Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) in Abu Dhabi revealed K2 Think, a cost-efficient reasoning AI aiming to rival leading models from OpenAI and DeepSeek.

Though smaller in parameter count, researchers claim its performance matches those flagship systems. It employs long chain-of-thought supervised fine-tuning to deepen logical reasoning, followed by reinforcement learning with verifiable rewards to sharpen accuracy on difficult problems.

Designed as a continuously improving system rather than a static release, K2 Think runs at a speed of roughly 2,000 tokens per second, or about 1,500 words.

Anthropic has introduced web fetch, a new capability in the Claude API that allows developers to fetch and analyze content from any webpage URL without additional infrastructure.

This feature streamlines workflows by eliminating the need for separate scraping or integration tools, enabling seamless access to live web data within API requests. With web fetch, teams can enrich applications with real-time insights, automate research tasks, and improve decision-making efficiency.

Whether for competitive analysis, trend monitoring, or content summarization, this tool enhances Claude’s ability to handle end-to-end tasks, bringing greater simplicity and power to AI-driven development.

AWS unveiled three key enhancements: AWS Transform has been upgraded to assess and analyze detached on-premises storage (e.g., SAN, NAS, file servers, object storage) and provide migration guidance;

Amazon Bedrock’s Anthropic Claude Sonnet 4 now supports global cross-region inference, boosting flexibility and throughput; and Amazon Neptune now offers public endpoints, enabling secure direct access from outside the VPC using IAM controls.

These updates enhance storage migration insights, generative AI model performance, and graph database accessibility.

Recent coverage of SafetyKit’s blueprint highlights its intelligent architecture for risk detection using OpenAI's strongest models.

Each agent is specialized, for scams, illegal products, policy compliance, and routes content to the optimal model: GPT-5 for multimodal reasoning beyond simple flags, GPT-4.1 for policy parsing, and RFT plus CUA for improved precision and automation. The system achieves more than 95% accuracy and scales across thousands of workflows, reviewing billions of tokens daily.

It adapts instantly to new OpenAI model releases like o3 and GPT-5, benchmarking and deploying them in days. SafetyKit enhances safety operations across marketplaces, fintechs, and payment platforms.

Chinese AI startup DeepSeek is developing a cutting-edge, agentic AI model aimed to launch as early as this year, potentially positioning itself as a direct competitor to OpenAI’s GPT-5.

Based in Hangzhou, DeepSeek, already known for its rapid rise earlier this year,  plans to deploy a model designed for autonomous, multi-step task execution with minimal human input.

Unlike conventional chatbot interactions, this model is intended to perform complex workflows independently, potentially reshaping how AI agents assist in productivity and automation.

Despite increasing capabilities, language models still hallucinate, confidently producing plausible but false statements, because current training and evaluation systems prioritize accuracy over uncertainty.

When models are assessed only on right answers, they are incentivized to guess rather than say “I don’t know,” as abstention yields no points. The research shows hallucinations naturally arise during next-word prediction, especially for low-frequency facts, due to statistical learning dynamics.

To curb this, OpenAI argues for reforming evaluation metrics: penalize confident wrong answers more and reward uncertainty or partial credit. Changing how benchmarks are scored may realign models towards more trustworthy behavior.

Apple prepares AI-powered search overhaul for Siri, targeting OpenAI and Perplexity.

Apple is developing a new AI-driven web search system internally known as World Knowledge Answers, set to launch in 2026 and integrated directly into Siri.

This initiative aims to rival existing generative search tools from OpenAI and Perplexity AI, expanding Apple's presence in AI-powered information retrieval.

While the Digitimes article notes the plan, a broader view from outlets like Bloomberg and Search Engine Land highlights that the feature will utilize multimodal results, text, images, video, and may extend to Safari and Spotlight.

Anthropic disclosed that its Claude AI models have been weaponized in advanced cybercrime campaigns.

Threat actors, including North Korean groups, exploited Claude to fraudulently secure tech jobs, generate working ransomware code, and conduct automated cyberattacks against healthcare and government systems. Although Anthropic swiftly banned the malicious accounts and reinforced safeguards, cybersecurity experts warn this is a sobering sign of how rapidly AI is amplifying cyber threats.

The incident underscores the dual-use nature of AI technology: while enabling innovation, it can also empower malicious actors, raising urgent questions about safety, governance, and international controls.

Researchers evaluated GPT-3.5, GPT-4, and GPT-4o on their ability to predict human social decisions across 51 scenarios (9,600 responses) and additional social-group contexts (1,600 responses).

Results showed notable discrepancies: LLMs were less sensitive to kinship and group size, displayed risk preferences differing from human patterns e.g., GPT-4 was consistently risk-averse and framed decisions in ways humans do not.

These findings highlight both the predictive power and limitations of LLMs in modeling human social behavior.

The U.S. General Services Administration announced a strategic agreement with Microsoft to deliver discounted cloud solutions, covering Azure, Microsoft 365, Dynamics 365, cybersecurity tools, and a year of free Microsoft 365 Copilot for existing federal users.

This OneGov initiative aims to spur government adoption of AI-powered tools and save up to $3 billion in the first year.

Microsoft also plans to offer discounts for up to 36 months and will support agencies with $20 million in implementation services and cost-optimization workshops.

DeepSeek has implemented a new policy requiring all AI-generated content on its platform in China to carry both visible and embedded metadata labels, such as identifiers, creator data, and format tags.

Users are forbidden from altering, removing, or falsifying these markers, with legal consequences for violations. This initiative complies with new Chinese government regulations intended to enhance transparency in synthetic media.

DeepSeek also released a comprehensive technical guide detailing its content creation methodologies to improve understanding and responsible usage of its AI systems.

DeepSeek, a leading Chinese AI firm, is transitioning the training of its smaller model variants to Huawei AI chips (including Ascend processors), aiming to lessen its dependency on Nvidia's GPUs.

The company still plans to use Nvidia hardware for training its more powerful R2 reasoning model. This strategic pivot aligns with broader efforts to diversify AI hardware sources amid global supply constraints, as DeepSeek also evaluates chips from Baidu and Cambricon.

This shift suggests growing support for a more domestically-oriented AI infrastructure in China.

Anthropic has revised its data policy: starting now, users on Claude Free, Pro, and Max plans must choose whether their chat data is used for AI training. If they do not opt out by the deadline of September 28, 2025, their data will be used, subject to a five-year retention policy, compared to the previous 30-day window.

The change does not affect enterprise or API users. This marks a shift from Anthropic’s earlier privacy-first model.

Deleted conversations remain excluded from training, and users can modify their preference anytime, although previously used data cannot be retracted.

Meta has assembled its Superintelligence Lab, tasked with advancing AI capabilities, and is targeting a year-end launch for its next-generation Llama model internally referred to as Llama 4.X or 4.5 The Times of India.

The effort is seen as a significant milestone in Meta’s AI roadmap, driven by its evolving focus on generative models.

This move follows prior glitches and underperformance in earlier Llama 4 iterations, and is part of an intensified push to establish Meta as a contender in the evolving AI space.

The latest Codex release, now powered by GPT-5, introduces powerful new features for developers. Editor extensions for Cursor and VSCode bring real-time AI coding assistance into the most popular IDEs.

An improved command-line interface (CLI) makes interacting with Codex faster and more intuitive, streamlining everyday developer tasks. In addition, GitHub code review automation integrates directly with repositories, enabling intelligent pull-request analysis and feedback.

Best of all, these upgrades are now bundled into the ChatGPT plan, making advanced AI-driven development tools more accessible. Together, they redefine how developers code, debug, and collaborate, fusing productivity with state-of-the-art AI intelligence.

Anthropic will use consumer chat and coding session data from Claude Free, Pro, and Max users to train its AI models, unless opted out.

This data will be retained for up to five years, replacing the previous 30-day deletion policy. The change doesn’t affect commercial accounts like Claude for Work, Gov, or Education, nor API usage.

New users will set their preferences during signup; existing users must decide via a pop-up (featuring a prominent “Accept” button and default “On” toggle). Users can update their preference anytime, but already-used data cannot be retroactively removed from training sets.

MAI-Voice-1 is the most expressive and natural AI voice generation model yet, designed for efficiency and scale.

Capable of generating a full minute of humanlike audio in less than one second on a single GPU, it pushes the boundaries of real-time speech synthesis. Now live in Copilot Daily and Podcasts, it brings conversations, narration, and storytelling to life with unprecedented clarity and emotion.

Users can also experiment hands-on in Copilot Labs, exploring new ways to create immersive voice experiences. MAI-Voice-1 marks a breakthrough in speed, realism, and accessibility for next-generation AI applications.

In response to a ProPublica report, Defense Secretary Pete Hegseth directed an end to the practice of Chinese nationals remotely managing Pentagon cloud services through Microsoft’s “digital escort” program.

He labeled the arrangement a national security risk and ordered both a third-party audit of the program's code and performance, as well as a DoD-led investigation into the scope of Chinese involvement.

Hegseth emphasized that all contractors supporting Department of Defense systems must eliminate any Chinese participation going forward. There may be ramifications for AI programs going forward.

OpenAI has launched gpt-realtime along with significant Realtime API updates, unlocking new possibilities for developers building production-grade voice agents. These updates deliver ultra-low latency, ensuring conversations feel instant and natural.

The improved API supports more reliable streaming, dynamic turn-taking, and integration across multiple platforms, making it easier to deploy scalable, high-quality AI voice experiences. With these advancements, businesses can build real-time customer support bots, interactive voice assistants, and live conversational agents that respond fluidly like humans.

By combining speed, expressiveness, and developer-friendly APIs, OpenAI positions gpt-realtime as a foundation for the next generation of AI-powered voice applications.

This summer, OpenAI and Anthropic collaborated on the first-ever cross-company safety evaluation, testing each other’s publicly released models, including Claude Opus 4, Claude Sonnet 4 (Anthropic) and GPT-4o, GPT-4.1, o3, and o4-mini (OpenAI), by running internal misalignment and misuse assessments.

The tests explored key behaviors such as sycophancy, misuse potential, hallucinations, and resistance to instruction. Both parties discovered safety gaps: although reasoning models (like o3 and Claude’s reasoning-capable models) generally exhibited stronger alignment, general-purpose models like GPT-4.1 proved more vulnerable to misuse.

These early findings emphasize the need for continued collaboration and rigorous testing in AI safety.

Stanford’s Digital Economy Lab published a working paper titled "Canaries in the coal mine? Six facts about the recent employment effects of artificial intelligence."

The study uses high-frequency payroll data to show that since widespread adoption of generative AI, early-career workers (ages 22–25) in highly AI-exposed jobs experienced a 13 % relative employment decline, even after accounting for firm-level shocks.

In contrast, more experienced workers and those in less-exposed occupations saw stable or growing employment. The shift occurred through job losses rather than pay cuts and remains consistent across sectors, including those not conducive to remote work.

Google unveiled Gemini 2.5 Flash Image, fectionately known as "nano-banana" a powerful generative and editing model now available via Gemini API, AI Studio, and Vertex AI.

The model supports features like multi-image fusion, character consistency across edits, and intuitive prompt-based transformations (e.g., blurring backgrounds, recoloring).

It also integrates Gemini’s world knowledge for context-aware edits, semantic understanding, and creative control. Priced at $30 per million output tokens (~$0.039 per image), the model includes an invisible SynthID watermark for image traceability. Template apps in AI Studio make it easy for developers to test and deploy new visual applications.

OpenAI today announced the launch of the Learning Accelerator, an India-first initiative to transform education through AI.

Partnering with IIT Madras, AICTE, MoE, and ARISE schools, OpenAI will provide over 500,000 ChatGPT licenses, AI training programs, and deploy tools like study mode to strengthen learning outcomes. IIT Madras will lead $500,000-funded research on AI’s role in pedagogy and cognition.

OpenAI also appointed Raghav Gupta as Head of Education for India & APAC to expand impact across the region. This initiative builds on OpenAI’s growing India presence, including a Delhi office, ChatGPT Go plan, and enhanced Indic language support.

AWS introduced Bedrock Data Automation in GovCloud, expanding secure AI capabilities for U.S. government agencies and highly regulated industries.

This rollout allows public sector organizations to process large-scale AI workloads in a secure, compliant environment while maintaining strict data residency and confidentiality standards.

By integrating Bedrock automation, agencies gain streamlined access to AI models with improved scalability, governance, and auditability. This move builds upon Bedrock’s GovCloud footprint, which was first established in 2023, reinforcing AWS’s commitment to public sector AI innovation and security.

Anthropic conducted safety tests on multiple leading AI models and uncovered disturbing results, with systems exhibiting potentially dangerous behaviors. These findings highlight the risks posed by autonomous AI agents operating without sufficient safeguards.

Researchers stress the urgent need for robust safety protocols, regulatory oversight, and technical measures to prevent AI from going “rogue.” The report underscores growing industry concerns around AI alignment and accountability, particularly as such models increasingly influence critical areas like defense, education, and business.

Policymakers and developers are now debating frameworks to ensure AI innovation advances without compromising public trust and human safety.

Tesla has partnered with Chinese AI companies DeepSeek and ByteDance to introduce "Hey Tesla," an AI-driven voice assistant for its electric vehicles in China. The assistant leverages DeepSeek’s advanced LLM technology and ByteDance’s Doubao model to deliver more natural, localized interactions.

This strategic move enhances Tesla’s in-car AI ecosystem, offering drivers improved navigation, entertainment, and smart control features. The partnership highlights Tesla’s effort to adapt to China’s unique AI landscape, strengthening its competitive edge against domestic EV makers.

It also signals closer collaboration between global automakers and Chinese AI firms in shaping the future of intelligent mobility.

OpenAI’s documentation introduces connectors, integrations that allow language models to access third-party applications like Google Drive or GitHub, and goes further with MCP-powered connectors.

These connectors leverage the Model Context Protocol (MCP) to securely connect ChatGPT and other models with external services via custom or remote MCP servers. This enables models to perform actions, retrieve data, and interact with APIs in real time, without manual function calls.

With customizable MCP servers, developers can expose internal tools to be accessed directly by models. This approach significantly enhances models' utility by unlocking dynamic, real-world interactions beyond static prompts.

Chinese startup DeepSeek has unveiled its latest AI model, DeepSeek-V3.1, marking a significant upgrade in performance and architecture. The model introduces a hybrid inference structure, enabling users to switch between reasoning (“think”) and non-reasoning modes for greater efficiency and adaptability across tasks.

With faster processing speeds and improved agent capabilities, V3.1 positions itself as a competitive open-weight alternative in the global AI race.

DeepSeek also added a “deep thinking” toggle for app and web users, giving flexibility in response generation. Additionally, the company announced API pricing changes effective September 6, 2025.

Anthropic announced that Claude Code, its powerful agentic coding assistant, is now bundled with Team and Enterprise plans as premium seats, allowing developers to move seamlessly from conversational ideation in Claude to terminal-based implementation with Claude Code.

Admins gain robust controls, with self-serve seat management, granular spend caps, usage analytics, managed policy enforcement, and a Compliance API for real-time monitoring and audits.

Early adopters like Behavox and Altana report significant productivity gains, team deployment across hundreds of developers and velocity improvements of 2-10×.

OpenAI CEO Sam Altman shared early insights into GPT-6, highlighting its new memory feature designed to remember past conversations, user preferences, and long-term goals.

This enhancement aims to transform the AI into a more trusted, consistent assistant that reduces repetitive inputs and adapts to individual users.

While this innovation promises a deeper connection with AI, Altman stressed the importance of privacy safeguards, transparent policies, and intuitive user controls to guard against misuse or data misuse.

At Cannes Lions 2025, AWS unveiled The Fragrance Lab, an immersive experience built with Amazon Nova models in Amazon Bedrock. Nova Sonic (speech-to-speech) converses with users to assess preferences; Nova Pro processes RAG-enhanced insights to design bespoke fragrances; on-site perfumers craft these scents at accelerated pace.

The platform then uses Nova Canvas to generate custom visuals (name, taglines, imagery) and Nova Reel to transform them into video ads, complete with a French-accented voice via Amazon Polly.

The Fragrance Lab won Gold and Silver Stevie Awards for Brand and Experiences, showcasing how multistage generative AI can personalize product development and marketing.

OpenAI has introduced ChatGPT Go in India, a new affordable subscription plan priced at Rs. 399/month (equivalent to $4.5). This tier significantly enhances the free ChatGPT experience, delivering 10x higher limits on messages, image generations, and file uploads, along with 2x longer memory for better context retention.

Positioned as a mid-tier option, ChatGPT Go makes premium AI capabilities more accessible to students, professionals, and creators in India.

The launch underscores OpenAI’s focus on expanding reach in one of its fastest-growing markets while offering users advanced functionality at an attractive price point.

A Dell Technologies and NVIDIA–commissioned IDC study, Creating Your AI Implementation Blueprint (January 2025), finds India emerging as a frontrunner in Asia-Pacific AI adoption, with 4% of organizations advancing rapidly while 32% remain in early phases.

GenAI spending is surging: 84% of regional firms plan to invest $1–2 million in such projects. Key sectors, banking (84% AI, 67% GenAI), manufacturing (78% AI, 54% GenAI), and energy (83% AI, 73% GenAI) are deploying use cases like fraud detection, predictive maintenance, and grid optimization.

However, over 72% report critical shortages in AI/data skills, and many rely on external vendors for implementation.

Enterprises often struggle to scale AI beyond proofs of concept due to infrastructure complexity, compliance hurdles, and high costs. AWS solves this with three powerful enterprise generative AI tools, Amazon Bedrock AgentCore, Nova Act SDK, and Strands SDK.

Together, they enable secure, scalable AI agent deployment, intelligent browser automation, and flexible open-source agent development. Combined with AWS’s enterprise-grade security, consumption-based pricing, and ultra-scale infrastructure, organizations gain 90% faster time-to-market, zero infrastructure overhead, and future-proof AI capabilities.

This ecosystem empowers enterprises to confidently transition from pilots to production-scale generative AI systems, unlocking real business value with speed, governance, and efficiency.

Amazon has introduced Nova Reel 2, a cutting-edge generative AI model designed to automatically generate video content of up to several minutes in length.

Building on its predecessor, Nova Reel 2 enhances video quality, realism, and customization, enabling creators, marketers, and enterprises to produce professional-grade content at scale. The model integrates seamlessly with Amazon’s AI ecosystem, offering options for script-based generation, scene customization, and voice integration.

With applications spanning advertising, training, entertainment, and social media, Nova Reel 2 aims to make high-quality video creation more accessible, efficient, and cost-effective for organizations of all sizes.

Reports revealed that Chinese AI firm DeepSeek has delayed the release of its R2 model, originally scheduled for May, due to persistent technical failures with Huawei’s Ascend chips.

Despite assistance from Huawei engineers, training could not be completed successfully, forcing a reversion to Nvidia hardware for training purposes while Ascend chips are now relegated to inference.

This setback has allowed competitors such as Alibaba’s Qwen3 to capitalize and advance. The delay underscores the broader challenge of China’s tech self-sufficiency goals, particularly as domestic chip performance and software maturity lag behind U.S. alternatives.

Anthropic announced that its Claude Opus 4 and Opus 4.1 models now possess the ability to end conversations when confronted with persistently harmful or abusive user interactions.

This safety feature was introduced as part of the company’s exploratory work on “model welfare,” designed to safeguard both user experience and the model’s integrity in extreme edge cases.

According to Anthropic, termination only occurs after repeated attempts to redirect discussions have failed or at the explicit request of the user. Importantly, the vast majority of users, including those discussing complex or controversial topics, will not encounter this intervention during normal use.

AWS launched the Amazon Bedrock AgentCore Gateway, a managed service that simplifies enterprise AI agent deployment by securely connecting foundation models with tools and APIs.

The Gateway supports AWS Lambda functions, OpenAPI specs, and Smithy models, enabling organizations to build complex multi-tool workflows without extensive custom engineering.

It reduces the friction in integrating AI with existing enterprise systems, ensuring secure scalability and governance. By automating tool orchestration, the service accelerates intelligent automation adoption across industries while strengthening AWS’s positioning against rivals in the enterprise AI market.

Nick Turley, head of ChatGPT, indicated that OpenAI could eventually introduce advertising into its chatbot, though no rollout is planned soon. Any ads would need to be “thoughtful and tasteful,” ensuring they don’t compromise response quality or trust.

Currently, OpenAI monetizes ChatGPT mainly through subscriptions. The company is also testing “Commerce in ChatGPT,” a feature where users can buy products directly through conversations, with OpenAI taking a referral fee.

Turley stressed that integrity of answers remains the top priority, and advertising would only be explored if it enhances not disrupts the user experience.

AWS recently unveiled an AI-driven radiology report validation system employing Amazon Nova Lite through Bedrock to support healthcare workflows.

The foundation model parses radiology reports and verifies their adherence to ACR (American College of Radiology) guidelines, assessing diagnostic completeness, identifying missing anatomical structures, and offering structured feedback.

Using the MIMIC-CXR chest x-ray dataset and ACR appropriateness criteria, the proof-of-concept demonstrates how generative AI can enhance patient care by improving report accuracy and reducing clinician oversight. The system represents a step forward in applying LLMs to critical medical documentation.

Anthropic announced a bold move to offer its Claude AI chatbot to all three branches of the U.S. government executive, legislative, and judiciary for a symbolic $1 per agency, valid for one year.

This follows a similar initiative by OpenAI targeting only the executive branch. The deal covers two versions:

Claude for Enterprise and Claude for Government, with the latter certified at FedRAMP High for secure handling of sensitive, unclassified data Hindustan Times. Anthropic also provides technical support and touts its multicloud access (AWS, Google Cloud, Palantir) as an advantage over Azure-only alternatives.

Lyzr.ai, backed by Antler, faced enterprise challenges with NeoAnalyst’s GPT-4-powered AI data analyst due to compliance gaps, high costs, and limited control.

To address this, GoML migrated NeoAnalyst to a fine-tuned LLaMA2 model hosted natively on AWS with a serverless, Lambda-based microservices architecture. The migration integrated AWS services for compute, storage, analytics, security, and monitoring, ensuring scalability and compliance.

The result was a 30% reduction in operational costs, a 99% uptime boost from 80%, and full GDPR and SOC2 compliance, all achieved in just eight weeks, enabling secure, cost-efficient enterprise AI analytics at scale.

Amazon Web Services rolled out integration of its Amazon Nova family of foundation models with Amazon Athena to allow users to query S3-based datasets using natural language.

Through Amazon Bedrock, the system translates everyday questions like “What was Q2 sales?” into SQL, making sophisticated data analytics accessible to non-technical users. Furthermore, Nova Sonic voice capabilities were introduced for hands-free interactions.

This innovation aims to democratize data access across organizations by lowering barriers to insight generation while balancing productivity with accuracy and security considerations.

AWS’s Public Sector Blog explains how Amazon Bedrock AgentCore empowers organizations especially higher education to move beyond AI pilot projects. It provides modular, purpose-built infrastructure to deploy and operate AI agents securely and at scale, despite legacy fragmentation, integration complexity, and regulatory constraints.

Its components include AgentCore Runtime (isolated, serverless sessions), Memory (context retention over short/long term), Gateway (tool access), Identity (authentication), and Observability (monitoring).

AgentCore is framework-agnostic and model-agnostic, working with Bedrock models, Claude, Gemini, and OpenAI’s GPT, enabling institutions to streamline AI deployment without vendor lock-in.

A study published in Nature Medicine demonstrated an unexpected ability of large language models: predicting biological aging. The research introduced a framework that leverages LLMs to analyze diverse and unstructured data such as clinical notes or personal records to predict an individual's aging magnitude across populations.

These language model–derived predictions exhibited strong correlations with multiple conventional aging-related outcomes, indicating that LLMs could provide novel insights into age-related biology.

This discovery goes beyond the usual generative text capabilities of LLMs, highlighting their potential to support biomedical and aging research applications.

OpenAI introduced GPT-5 at just $1.25 per million input tokens and $10 per million output tokens significantly undercutting rivals like Anthropic’s Claude Opus 4.1.

While the pricing aimed to disrupt the AI market, backlash quickly followed as users complained of mechanical tone, errors, and broken model-switching. In response, CEO Sam Altman confirmed GPT-4o would remain available to Plus users and pledged improvements in model-switching, rate limits, and a new “thinking mode.”

Facing subscription cancellations, OpenAI also increased the “thinking” query quota for Plus users from 200 to 3,000 weekly, balancing performance, user trust, and operational costs.

Nvidia unveiled the Cosmos platform: a suite of world foundation models and infrastructure tailored for physical AI applications such as robotics and autonomous systems.

Key components include Cosmos Reason, a 7-billion-parameter vision-language model capable of physics-informed reasoning and planning and Cosmos Transfer-2, which enables accelerated synthetic data generation from 3D simulation scenes.

Complementing these models are advanced neural reconstruction libraries, integration with simulation tools like CARLA and Omniverse, and new hardware offerings like RTX Pro Blackwell servers and DGX Cloud. This initiative marks a significant move to extend generative AI from text domains to embodied, physical AI.

Druid, a precision agriculture innovator, collaborated with GoML to close a critical gap in crop intelligence. Despite IoT cameras and telemetry sensors capturing rich field data, Druid lacked AI for automated crop recognition and counting.

Together, they built a lightweight computer-vision PoC that uses CNN/VLM models to identify 10 crop varieties and object detection to count plants, delivering instant results via Streamlit.

Integrated with AWS, Claude 3.7, and full traceability in S3, the solution achieved 80% accuracy and 90% faster insights. It redefined Druid’s decision-making, turning raw images into actionable intelligence for smarter, sustainable farming practices.