Anthropic's pre-release Claude Mythos model was accessed by unauthorized users despite the company's public claims that it was too dangerous to release. The breach exploited information from a separate Mercor data company breach combined with insider knowledge to guess the model's online location. The incident undermines Anthropic's positioning around AI safety.
Safety
Anthropic’s Mythos breach was humiliating
Anthropic's unreleased Claude Mythos model was breached despite company claims it was too dangerous to release, undermining its safety-first positioning.
Thursday, April 23, 2026 12:00 PM UTC2 MIN READSOURCE: The VergeBY sys://pipeline
Tags
safety
/// RELATED
Strategy5d ago
At his OpenAI trial, Musk relitigates an old friendship
Musk's lawsuit testimony reveals his 2015 split with Larry Page over AI safety philosophy — Page accepted existential risks as the price of progress, prompting Musk to launch OpenAI as a safety-first counterweight.
Research5d ago
Automated detection of pediatric congenital heart disease from phonocardiograms using deep and handcrafted feature fusion
Researchers combine deep learning with handcrafted audio features to enable automated detection of congenital heart disease in children from phonocardiogram recordings, paving the way for accessible non-invasive cardiac screening.