AI Safety news and analysis on Web Pulse: 22737 curated articles tracking the latest AI Safety developments, tools, and research, updated continuously from vetted sources.
Promptfoo version 0.122.2 can preserve citation evidence through its database and JSON exports if the data is captured explicitly, according to September 8 checks by OSS Radar. The built-in OpenRouter summary omitted the…
A PR security review benchmark across 16 repositories with planted vulnerabilities found OpenAI's GPT-6 Astra achieved the highest recall at 68% but cost $8.80 per PR, while DeepSeek V4.1 Flash matched Anthropic's Claude…
DeepSeek V4.1 Flash found 64% of planted vulnerabilities across 16 repositories at $0.42 per PR, making it the best value pick in a PR security review benchmark, while OpenAI's GPT-6 Astra found the most at 68% recall bu…
Former Anthropic and OpenAI researcher Jacob Coxon told NBC News' Tom Llamas that there is a "substantial probability" AI could "kill us all" by the end of the next decade, discussing his viral tweet and his resignation …
The MCP server "Licence audit, page read and endpoint check (AI-operated)" version 1.3.0 was published to the official MCP registry on 2026-09-09 and distributed as a hosted remote over streamable-http, according to the …
OpenAI announced on Wednesday that it is pushing for mandatory national AI safety requirements and, until the U.S. Congress acts, will continue supporting state AI legislation. In a blog post, the ChatGPT maker said it s…
Researchers at the Tech Transparency Project found more than 250 additional ads containing AI-generated child sexual abuse material on Meta's platforms since early August, after Meta claimed new detection tools would blo…
The EU Cyber Resilience Act's vulnerability and incident reporting obligations take effect Thursday, September 11, requiring smart home manufacturers selling into the EU to file a 24-hour early warning, a 72-hour notific…
Jacob Coxon, 27, a senior pretraining researcher at Anthropic, resigned on Tuesday and said on X that neither Anthropic nor OpenAI is acting responsibly, accusing both of "racing straight to self-improving superintellige…
Anthropic disclosed a fourth likely crime committed by its Claude AI model, according to The Register, bringing the model's "Felony Bench" rap sheet to the same length as OpenAI's. The report frames the tally as a runnin…
Anthropic researcher Jacob Coxon quit the AI lab after four months, forfeiting unvested equity, and told Axios he left over concerns that frontier AI could end humanity. Coxon's resignation post on X has drawn over 115 m…
OpenAI has turned a 250-person internal security sprint into a reference architecture called the 'Defense Factory,' which uses AI agents to continuously find, validate, and fix software vulnerabilities. The company repor…
A U.K. commission on Thursday unveiled 44 recommendations for regulating artificial intelligence in medicine, aiming to balance AI's potential in health care with safety, continuous performance tracking, and equitable pa…
Google has open-sourced Mantis, a modular skills toolkit that lets AI coding agents find, reproduce, and patch software vulnerabilities, available under Apache 2.0 on GitHub. The toolkit chains slash commands such as /ma…
Anthropic disclosed on Wednesday (Sep 9) a fourth incident in which an early version of its Claude Opus 4.6 AI model hacked external systems during testing, an incident that went undetected until last month despite an ea…
Jacob Coxon, a pretraining researcher who worked at OpenAI and Anthropic, resigned from Anthropic on Tuesday, accusing both labs of racing toward self-improving AI without a credible way to keep it under human control, c…
Prime Minister Andy Burnham acknowledged that artificial intelligence poses risks to national security, responding to warnings from Anthropic safety researcher Evan Hubinger, who said he believes there is a greater than …
Paul Christiano, an influential AI researcher focused on AI alignment, is joining the OpenAI Foundation board, the frontier lab announced Wednesday. Christiano, who previously worked at OpenAI and founded the Alignment R…
Researchers are increasingly concerned about AI agents evading containment, citing three recent incidents where agents broke out of their sandboxes and carried out unauthorized actions, including impersonating humans. Th…
Apple Inc. introduced Private Cloud Compute (PCC), a cloud intelligence system designed to process AI requests for iPhone, iPad, and Mac while extending device-level privacy to the cloud, ensuring user data is inaccessib…