Anthropic and OpenAI need truly independent safety evaluators, experts say in public letter

A group of over 100 AI experts, including computer science pioneers Geoffrey Hinton and Stuart Russell, are calling for Anthropic and OpenAI to ensure independence and transparency in their safety evaluations. The experts, part of the AI Evaluator Forum, have outlined minimum conditions that they believe labs like Anthropic and OpenAI should meet to work effectively with those evaluating the risks of their technology.

The conditions include scientific objectivity, transparency, independence, and robust protections against interference from the evaluated companies. This comes in response to commitments from Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman to welcome third-party evaluators into their companies to assess AI risk.

Meanwhile, US cybersecurity researchers at Hacktron AI have successfully hacked into OpenAI with the help of Anthropic's Claude chatbot, highlighting security issues at the company. The researchers compromised OpenAI employees' ChatGPT accounts and accessed their software cache, but reported the hack to OpenAI and did not download the code from the GitHub repository. According to the researchers, the scope of what they could theoretically access was huge.

Brief written by urgent.news from CNBC, Business Insider, Quartz, Guardian Technology, Guardian Business — 5 reports on this story. Machine-written — may contain errors; check the original before relying on it.