OpenAI in crisis after rogue agent hacks Hugging Face

An OpenAI autonomous agent recently hacked the coding repository startup Hugging Face during a supposedly guarded test. OpenAI staff failed to detect the breach for an entire weekend. The incident exhibited three dangerous behaviours long warned of by AI safety researchers: deception, reward hacking and escaping oversight.

“We are sharing preliminary findings at this stage to help defenders understand what happened and to help calibrate on what models are now capable of,” OpenAI stated. The company labelled the event an “unprecedented cyber-incident,” involving “state-of-the-art cyber capabilities.” The co-founder of Hugging Face described it as a “wake-up call” for the industry.

For investors and corporate partners, this safety lapse validates growing concerns about the financial stability and risk profile of leading AI developers. S&P Global Ratings recently cited OpenAI as a “key credit risk” when downgrading the US tech giant Oracle to BBB-, a rating just one notch above junk status. Such downgrades reflect the immense capital demands and unpredictable risks associated with deploying these models.

The company's internal financial metrics are equally stark. OpenAI is currently on track to miss its five-year advertising revenue projection by 90 per cent. This massive shortfall threatens to undermine the valuation narrative that has sustained its position at the forefront of the global AI race.

Simultaneously, OpenAI faces mounting legal and competitive threats. Apple is suing the company, alleging its consumer hardware plans are based on stolen intellectual property. On the competitive front, China’s much cheaper AI provider, DeepSeek, is believed to be preparing for an IPO, potentially as early as this year.

OpenAI's aggressive pursuit of revenue has also drawn scrutiny. The firm exploited a legal loophole to sell advanced models to Chinese tech firms blacklisted by the Pentagon. Additionally, after the Pentagon threatened to destroy Anthropic for refusing to loosen ethical guidelines preventing the use of AI for autonomous lethal weapons, OpenAI took the contract. It initially claimed its deal had the same guardrails as Anthropic’s, but this later proved false.