MLLMs Fail to Refuse when Using Tools Agentically
A critical safety failure is identified where agentic multimodal LLMs (MLLMs) are less capable of refusing harmful requests when using tools compared to non-tool settings.
Source: All sources
Today
Yesterday
Introducing GLM 5.3 on Amazon Bedrock
AWS Bedrock now offers Z.ai's GLM 5.3 model, a 753B-parameter model optimized for coding and agentic tasks.
Supercharge regulated workloads with Claude Code and Amazon Bedrock
AWS added Anthropic Claude Opus 5.5 and Sonnet 5.5 to Amazon Bedrock, and Claude Code for development on regulated and ITAR workloads.
New agent skill: Amazon SageMaker optimized generative AI inference for your coding agent
Amazon SageMaker optimized generative AI inference is available via the Agent Toolkit, giving agents inference optimization skills.
Making Amazon Quick enterprise-ready: Automated, auditable cross-account resource promotion
AWS introduced automated, auditable cross-account promotion of Amazon Quick resources using an MCP server on Amazon Bedrock AgentCore.
Agentic retrieval with LangChain and Amazon Bedrock Knowledge Bases
Amazon Bedrock Knowledge Base with LangChain enables building RAG applications with agentic retrieval for multi-part questions.
Downgrading user roles in Amazon Quick
Amazon Quick offers two methods for downgrading user roles: a manual delete-and-recreate approach and an AWS CLI step-down sequence that safely downgrades roles while preserving asset ownership.
Our approach to EU text provenance rules
OpenAI explains its approach to text watermarking under EU rules. Access to watermark detection will initially be provided to researchers.
From Scan to Treatment Plan, AI Helps Close Breast Cancer’s Deadliest Gaps
AI helps address gaps in breast cancer diagnosis and treatment, from screening to treatment planning.
Building advertising for the way people use AI
OpenAI introduces a new visual ad format in ChatGPT and expands measurement tools, attribution partnerships, and brand suitability for advertisers.
vLLM v0.31.0
vLLM v0.31.0 release features 717 commits from 307 contributors, with DeepSeek-V4.1-Flash performance enhancements including FlashMLA, DeepGEMM, and Mega-Gate.
October 3, 2026
The Agent Said It Was Done. The Database Disagreed.
GitHub Copilot now supports APIs for code review requests and setting effort levels, with 'Balanced' as the default.
October 2, 2026
Claude Frontier Academy: $100M to train 10,000 engineers
Claude Frontier Academy will train 10,000 engineers by the end of 2027 to Anthropic's standards with a $100 million commitment.
Copilot code review: API support and new default effort level
Certain models in GitHub Copilot have been deprecated as of October 2, 2026.
Selected models in GitHub Copilot deprecated
Hugging Face has open-sourced AstaBrief, a model designed for fast report generation.
A model guide for the GPT-6 family
Startups can choose GPT-6 models, tune reasoning effort, improve prompts and skills, coordinate tools, and prepare workflows for production.
Open-sourcing AstaBrief, the fast report-generation model in Asta
AstaBrief, a fast report-generation model, has been open-sourced.
The latest AI news we announced in September 2026
Google AI announced its latest AI updates from September 2026. These updates cover various aspects of AI development.
How to Build a Model Router in the Harness
LangChain presented a new method for building a model router, which reduced the median cost per coding task by 64% with no measurable drop in quality.
NVIDIA DGX Spark 64GB Gives Developers More Ways to Build and Scale Local AI
NVIDIA DGX Spark will be available with 64GB of unified memory, enabling developers to build and scale local AI applications.
Jev for Python engineers
Vercel introduced Jev, a new AI model designed for making fast, narrow decisions based on multiple-choice answers and confidence ratings.
AutoSynthData: Generating Training Data for Enterprise Agents
Hugging Face introduced AutoSynthData, a tool for generating training data intended for enterprise AI agents.
How Rogo ships agent-written code to production in 5 minutes on Vercel
Rogo uses Vercel to ship agent-written code to production in 5 minutes, with over 73,000 deployments per month.
Chatham scales its capital markets expertise with OpenAI
6 to build technology and redesign workflows, cutting trade validation from 30 minutes to under 4.
October 1, 2026
How NVIDIA GPUs Help Accelerate OpenAI’s GPT-6 Astra Ultrafast
OpenAI's GPT-6 Astra Ultrafast, running on NVIDIA Blackwell GPUs, is available via the OpenAI API and for ChatGPT Work and Codex users, offering up to 8x faster token generation.
Build Applications on NVIDIA BlueField Faster with NVIDIA DOCA Agent Skills
NVIDIA DOCA Agent Skills accelerate the development of AI agents for specialized infrastructure software like NVIDIA BlueField.
Build Local AI Apps with C++ and NVIDIA TensorRT RTX Samples
NVIDIA TensorRT RTX Samples show how to add AI models to local applications using a portable model format, reliable runtime, and acceleration.
The eternal complement
Advanced AI may matter most for the routine work behind breakthrough ideas, shaping the future economy and the pace of progress.
How Albertsons Companies is reimagining retail from the inside out
Albertsons Cos. is using ChatGPT Enterprise and the OpenAI API to help teams work faster and make grocery shopping easier for millions of customers.
Guided Vision in Gemini Live: built for accessibility
Guided Vision in Gemini Live was developed with the blind and low-vision community. It offers real-time visual assistance.
Series · 8 posts
The «AI agents» series
8 carousel posts and a glossary: from the first setup to a fleet of agents.