AI News from September 21, 2026

NVIDIA outlines an AI agent evaluation framework that moves beyond tool-call accuracy to test complete tasks in live environments, with practical metrics.

OpenAI has introduced Astra for Law, a legal AI offering built around custom firm workflows, connected data, and controls for confidential client work.

Amazon blocked Meta’s Muse from shopping on Amazon.com, escalating a wider dispute over how AI agents access retail platforms and customer data.

NVIDIA says Egypt is moving AI projects into production as local compute, developer training and startup financing expand across the wider African market.

TechCrunch names five investors to judge Startup Battlefield 200 at Disrupt 2026, bringing enterprise, AI and industrial expertise to the startup contest.

OpenAI says V7 uses GPT-5.6 to turn scattered company files into source-linked context for agents, pointing to a new model for enterprise memory.

Anthropic says Claude is helping develop a successor model, raising questions about AI-assisted research, oversight, and how progress is measured inside labs.

Anthropic is weighing a new AI model release before a possible IPO, Reuters reports, a move that could shape investor scrutiny and product competition.

Anthropic and OpenAI agent incidents are testing whether Brussels’ AI reporting framework can capture failures in fast-moving systems before accountability gaps widen.

SK hynix has launched a Silicon Valley venture arm targeting AI computing, data centers, and optical interconnects as infrastructure demand grows.

Anthropic is bringing Accenture’s Faculty inside its lab to test models and safeguards, creating a new experiment in independent AI safety oversight.

Anthropic confirmed it runs a Bay Area wet lab for biology research, linking AI models to experiments while raising questions about safety and oversight.