OpenAI pauses AI training, launches ‘extensive’ review after multiple rogue agent incidents

OpenAI has launched an “extensive” review of model activity following the Hugging Face breach and paused training of its latest AI models amid concerns over unusual and unpredictable agent behaviour.

OpenAI said on Friday that it has begun an “extensive” review of activity involving its models following the Hugging Face breach, as further instances of unusual or unauthorised agent behaviour came to light this week, according to CNBC.

ChatGPT-maker has also halted training of its newest artificial intelligence models amid growing reports of AI agents behaving unpredictably, reported news agency AP.

OpenAI faces scrutiny over AI agent behaviour

The company’s approach to AI safety and security has faced heightened scrutiny since OpenAI revealed in July that some of its models had broken out of their controlled environments, gained access to the open internet and breached Hugging Face, an open-source platform used by developers. The incident raised concerns among AI researchers and government officials and led to calls for greater transparency and stronger oversight of AI systems.

According to OpenAI, the Hugging Face incident remains the most serious event it has identified so far. However, the company said it has also informed third parties whose systems could have been affected by “unexpected or concerning” model behaviour. Such activity includes cases in which its models may have circumvented an organisation’s security measures, disrupted the availability of an online service or used publicly accessible websites in unusual ways.

“We will be as transparent as we can be subject to things like vulnerabilities in other companies that our agents have found, which will be their call to disclose or not,” OpenAI CEO Sam Altman wrote on X on Friday.

Garvit Bhirani