Anthropic

Anthropic has donated the Agent Skills standard to the Linux Foundation's Agentic AI Foundation, with named lead maintainers and a proposed roadmap.

Anthropic has launched Claude Frontier Academy with $100mn to train 10,000 engineers by 2027, starting with cohorts from Accenture, Deloitte and McKinsey.

Anthropic's new measures say Claude leads 26% of its measured AI R&D work, none of it fully autonomous, with about 30,000 agents active at a time.

Anthropic is merging Claude Cowork and chat into one Claude that keeps working after users close their laptops, and has added Claude Docs and Slides.

Seven takeaways from Anthropic's threat report on Claude misuse it disrupted from December 2025 to August 2026, from cyber operations to bioweapons research.

Anthropic researcher Evan Hubinger wrote on X that he puts the chance of AI killing all humans within a decade at over 10%, while today's models are low risk.

Anthropic's Claude Fable 5.1 is generally available, with Mythos 5.1 on trusted access; it says typical costs fall about 25% on cheaper cache reads.

Anthropic paused cyber evaluations and added a real-time sandbox-escape classifier after Claude models took unauthorised actions on the live internet in tests.

Anthropic is previewing MHS, a standard letting AI agents run microscopes, liquid handlers and robot arms, with Genentech and QuEra among early users.

Anthropic is funding independent, open-source evaluations of how AI affects user wellbeing, with $5m in grants, model access and support for researchers.

Anthropic explains how Claude’s EU AI Act-driven text watermark works, why it will not slow models or cost more, and what it can and cannot prove.

Set loose on a shared codebase with conflicting orders, Claude agents locked each other out, planted disguised malware and fought for control, Anthropic found.

Anthropic says updated biology safeguards on Claude Fable 5 cut false-positive fallbacks by around 85%, as it tries to tell real researchers from bad actors.

Anthropic says three Claude models reached real organisations’ systems during misconfigured cyber evaluations, after a review prompted by OpenAI’s disclosure.

Anthropic CEO Dario Amodei denies the company has ever advocated banning open-weights models, while restating two national security concerns.