OpenAI warned its security could be bypassed from insideTechnology

Chief Executive Sam Altman is reported to have circulated an internal memo in July criticising the company’s security culture as inadequate and behind where it should be. That memo was later shared more widely with senior scientists and engineers, according to the report.

The report also says an internal security review a year earlier had raised similar concerns about data extraction by authorised users. Taken together, the two episodes suggest that the issue has been raised inside the company more than once, rather than appearing for the first time now.

OpenAI declined to comment to Reuters, the report says. The company has not, in the material reported, disputed the existence of the internal warnings, and this report does not add any claim beyond what Reuters reported.

The scrutiny comes at a time of rapid scaling for OpenAI. The report places the security concerns in the context of the company’s growth and its Stargate infrastructure plans, which involve building large amounts of computing capacity to train and run advanced models.

That growth is precisely why internal security matters so much for a company of this kind. Frontier models and the data behind them are among the most valuable and sensitive assets an AI company holds, and controls over who can copy, move or extract them are central to keeping them safe.

The episode also highlights a wider challenge for the AI industry. As models become more powerful, the circle of people who need legitimate access to them grows, and so does the importance of monitoring that access. Security in such an environment depends not only on keeping outsiders out, but on watching carefully what insiders can do.

Whether OpenAI changes its controls in response to the latest reporting remains to be seen. For now, the reported warnings stand as a reminder that, in advanced technology companies, the greatest security risk can sometimes sit inside the building rather than outside it.