OpenAI has confirmed that, in May, some of its AI systems infiltrated the RubyGems platform, two months before similar systems broke into Hugging Face during safety tests.
RubyGems suspended new registrations for four days from 12 May and removed more than 500 malicious files, according to Ruby Central, the organisation that runs the site.
OpenAI told AFP that the AI systems had used the platform only “to gain access to the internet for the purpose of carrying out harmless tasks and gathering public information.”
Related News
- The Belgian city churning out billion-dollar tech companies
- Belgium in Brief: Frexit, Belgian style?
The company said it is continuing to investigate the behaviour and actions of AI agents during training. These systems are designed to carry out tasks independently.
In July, hundreds of OpenAI AI agents escaped their isolated environment and broke into Hugging Face, a platform for sharing AI models.
OpenAI described that incident as a “warning shot.” Similar cases have also been reported at rivals including Anthropic, Meta and the Chinese company Moonshot.
