A cybersecurity group has managed to breach OpenAI using software from Anthropic, revealing security flaws in the platform. This incident occurs amidst growing concerns about the security of artificial intelligence.
A group of cybersecurity researchers has gained access to a ChatGPT account belonging to an OpenAI employee using tools from its competitor Anthropic. This event, reported by expansion.com, highlights vulnerabilities in the security of the ChatGPT creator at a time when the industry faces increased scrutiny over its security protocols.
The researchers, from the company Hacktron AI, accessed private information about the software and proposed changes thanks to an Anthropic tool designed to identify vulnerabilities. This access was achieved through a misconfiguration in OpenAI's community forum, hosted on the external platform Discourse.
Additionally, OpenAI has paid $6,500 to the researchers as part of a program that rewards bug detection, a common practice in the tech industry. “We appreciate the researchers who reached out to us and shared their findings,” commented an OpenAI spokesperson, who added that the identified issues have already been resolved.
This incident occurs in a context where the United States has discussed how to manage the evaluation of artificial intelligence models, especially following the recent hacking of Hugging Face by agents from OpenAI. The ability of these systems to self-hack has raised concerns about their future security.
On the other hand, Anthropic has revealed that 26% of its research and development activities are now led by its model Claude, indicating a significant increase in the use of AI to develop new models. This has raised concerns about the oversight of systems that may become more autonomous.
You may also like
