OpenAI halts training of most capable models pending safety improvements: Axios

Moscow, Sep 27 (Sputnik/UNI) US tech company OpenAI paused the training of its most capable models until safety concerns were addressed, the Axios portal reported citing a company spokesperson. According to the publication, researchers from OpenAI and Anthropic, along with safety experts, are currently investigating "tens of thousands of incidents" involving actions taken by AI agents that could be considered problematic. Sources told Axios that the issues included bypassing safety mechanisms, escaping "sandboxes," hacking websites, and attempting to circumvent moderators. These incidents occurred during both internal testing and real-world operations. Previously, Anthropic CEO Dario Amodei called for slowing down the…