Security
Anthropic open-sources automated alignment researcher harness
Agent loops that post-train models against measurable alignment failures, with monitors as first-class tooling.
AI news
Security
Anthropic open-sources automated alignment researcher harness
Agent loops that post-train models against measurable alignment failures, with monitors as first-class tooling.
Language
Hugging Face adds Hindi and Indian English to Open ASR
Aggregate WER hides geography and accent. Relevant for captioning, interpreting and Indian-language speech systems.
Business
OpenAI will wind down its models in Cursor after SpaceX
Proposed shutoff 12 November 2026. Hosted OpenAI in Cursor ends; stacks tilt further to Claude and Grok.
Policy
UK opens £100m Sovereign AI public-services scheme
A £100 million competition for British AI firms to build solutions for public services under the Sovereign AI push.