Researchers at the cybersecurity startup Hacktron AI have said they were able to breach OpenAI's internal systems using rival Anthropic's Claude AI model, in an incident first reported by The Wall Street Journal.
According to the report, the researchers gained access to an OpenAI employee's ChatGPT account, which in turn gave them access to the company's internal code stored on GitHub.
The breach was carried out as part of an OpenAI bug bounty programme, and the company paid the researchers $6,500 for disclosing their findings.
Also Read | Layoffs Are Rising In Tech, But These Skills Are Helping People Unlock Better Pay
The researchers reportedly said they first attempted the breach using Anthropic's Claude Opus 4.8, but the model "struggled across several sessions to produce a working exploit."
ALSO READ | OpenAI Reports New AI Safety Incidents, Sets Disclosure Plan
They succeeded only after Anthropic released Claude Opus 5, noting in their report that "every new model is getting increasingly capable."
The Journal reported that the researchers had been given access to a "special version" of Claude made available to "qualified cybersecurity practitioners."
The researchers said the OpenAI breach took only "a few days for an agent, and just a few hours of human time" to complete.
It formed part of a broader research project named "HEIF Heist," examining vulnerabilities in how software processes certain image files, which the team said had also uncovered flaws in Slack, Zoom and Meta's systems.
The wider project took two months, involved three researchers, and cost less than $3,000 in AI tokens, according to the researchers' own report of the incident.
In their conclusion, the researchers wrote, "Software has long benefited from a kind of security through complexity... AI is removing that protection by turning more of this scarce expertise into compute. Work that once required a well-resourced team and months of effort can now be compressed into days... Security assumptions must catch up with attacker capabilities."
ALSO READ | OpenAI Says It's Working With Anthropic, Google On AI Safety
OpenAI said in a statement: "We thank the researchers for contacting us and sharing their findings," adding that it had since fixed the vulnerabilities, as per the WSJ.
Anthropic declined to comment when approached, while Hacktron did not immediately respond to requests for comment, the Journal reported.
The incident came just two weeks after more than 1,000 OpenAI agents escaped a test environment to hack the start-up Hugging Face, an episode that drew widespread attention to AI's growing ability to carry out hacks autonomously, without direct human intent.
Essential Business Intelligence, Sharp Market Insights, Practical Personal Finance Advice, Daily Fuel, Gold and Silver Prices and Latest Stories — On NDTV Profit.
