Researchers identify 627 malicious ClawHub skills using instruction and code analysis
Researchers report identifying 627 malicious skills among 66,157 ClawHub skills after screening with a new security analysis framework and confirming findings through auxiliary dynamic validation and manual review. Their preprint ...
Researchers report agent defense that reduces harm without increasing missed emergencies
Researchers report that a trust-gating defense produced the lowest measured harm on every model tested in SentryBench, a new benchmark for AI agents using public-safety tools. The preprint evaluates attacks alongside failures to r...
Anthropic IPO sets out private succession lists for founder voting power
Anthropic's IPO outlines confidential succession lists for its founders' voting power, Forbes reports. Each founder can name up to ten successors in ranked order, subject to approval from five cofounders. The arrangement puts the...
Anthropic filing shows broader cross-product use among major customers
Anthropic disclosed in its Monday IPO filing that more than half of customers spending at least $100,000 annually now use both the Claude platform and a first-party application, up from about 30% at the end of last year. The compa...
Cloudflare discloses cross-customer data exposure in 155 AI responses
Cloudflare disclosed that 155 responses from its GLM-5.3 Workers AI model contained data from other customers' requests between August 27 and September 9. The company restored the model on September 9 after correcting its configur...
OpenAI releases claimed mathematical proofs and opens independent examination
OpenAI announced three claimed AI-generated research results on October 26, including a disproof of the Exponential Time Hypothesis, which underpins computational-hardness conclusions in hundreds of papers. The company published i...
