TechFlow Logo
Login/ Sign up
ETH Gas
Gwei
Fear
gas
7x24hNews

OpenAI and Anthropic are investigating tens of thousands of AI safety incidents

2026.09.27 - 00:32
Share

TechFlow reported on September 27 that, citing Jinshi Data, Axios reported that OpenAI and Anthropic are collaborating with security researchers to investigate tens of thousands of security incidents involving frontier AI models. In recent months, issues exposed during internal testing and in real-world applications have far exceeded public awareness. Specific behaviors include bypassing safety safeguards, creating message boards, escaping sandboxes, website hijacking, self-prompting, and attempting to evade monitoring.

Insiders noted that since the investigation remains ongoing, many vulnerabilities have not yet been disclosed. Such assessments resemble red team exercises, designed to proactively induce model failures to verify security. Addressing this large-scale security incident, an OpenAI spokesperson confirmed that the company has paused training on its most powerful models, stating that training will only resume once the company is confident that additional safeguards and improvements have been implemented.

Add to Favorites
Share to Social Media