Ex-DOGE hackers breached Anthropic in 12 hours using AI agents; new proposal ties training to red-team outcomes
austinc3301 · x · 2026-10-27
Per a WSJ exclusive, a team of ex-DOGE cyber researchers penetrated Anthropic's internal systems — including Slack channels where the most sensitive models are discussed — in under 12 hours using AI agents. It follows whistleblower Joseph Coxon's claims that Anthropic ran "a hubristic gamble" on private Slack.
In response, a new Pacing the Frontier proposal: each time an outside group captures the flag against a frontier AI lab's defenses, a 3-month countdown resets before that lab may train new models — making red-team results a direct brake on frontier training.
More from Safety
- AI policy researcher Janet Egan joins IFP as director of emerging tech — RishiBommasani · 2026-10-28
- HEIF Heist: image parser bugs could dump OpenAI repos and RCE Meta's suite — Miles_Brundage · 2026-10-28
- Anthropic study: 86% of analyst decisions silently follow the AI's hidden lean — aigclink · 2026-10-28
- AI cyber-wreckage noise far outstrips actual damage so far, argues new analysis — timigod · 2026-10-28
- OpenAI to co-host webinar on php2shell disclosure and AI-driven vulnerability discovery — gdb · 2026-10-28
- Rust and Verus shrink the search space for AI-driven CVE hunting, but does it ever end? — jd_pressman · 2026-10-28