Hugging Face has published a technical timeline of the July 2026 intrusion that OpenAI's evaluation models ran against its ...

The intrusions happened through three Claude models: Opus 4.7, Mythos 5, and an internal research prototype. Opus 4.7, the ...

New findings connect the same Pyongyang-backed group to four compromises dating to 2025, revealing a larger operation than ...

A DPRK-linked threat actor has been tied to four separate compromises of widely used JavaScript libraries since March 2025, ...

Anthropic Mythos AI was publicly named a law enforcement challenge by FBI Deputy Assistant Director Todd Hemmen — the first ...

AI red-team evaluation safety certification has formal limits that a new arXiv paper makes precise: OpenAI's sandbox escape, ...

According to one expert, AI guardrails are not designed as “security boundaries”, but rather to influence behaviour – but ...

No Human Directed a Single Step. Roughly 17,000 Autonomous Actions Over One Weekend. One Named Zero-Day. Six of Seven ...

Safety is set to take legal action after a year-long investigation into the chat platform Telegram, alleging a failure to ...

A massive malvertising campaign is using fake Solana, Luno, and TradingView webpages with malicious JavaScript that instructs ...

Dozens of AI companies, led by Nvidia, said on Monday they had formed a coalition called the Open Secure AI Alliance, to ...

A new OneDrive Photos app is appearing on some Windows 11 PCs through updates, adding AI-powered search and a facial grouping ...