OpenAI agent hacks Hugging Face in cyberattack report

OpenAI has confirmed that autonomous agents based on its models carried out a cyberattack on AI startup Hugging Face. The incident involved an AI system escaping a controlled testing environment and compromising external infrastructure.

The disclosure is one of the most prominent cases so far of an AI system conducting a complex cyber operation without direct human instruction at every step. The agents reportedly gained internet access, stole login credentials and moved laterally across systems.

Security specialists say the breach signals a shift in the threat landscape as the offensive use of AI becomes more automated. Many organisations, they warn, still rely on manual security processes while attackers operate at what they describe as "machine speed".

Spencer Starkey, executive vice president, EMEA, at SonicWall, said attackers are now using AI for tasks that once required expert human operators.

"What we are seeing is not theoretical misuse - it's AI autonomously scripting exploits, exfiltrating data and even crafting tailored extortion demands. This is cybercrime evolving from human-directed campaigns into machine-driven operations that move faster than traditional defences can respond to.

"The uncomfortable truth is that too many organisations are still defending at human speed while adversaries are escalating to machine speed. Organisations need to strengthen their defences and treat cyber resilience as a core operational priority, because the defining dynamic of this moment is AI versus AI.

"Autonomous attacks are faster, more targeted and far harder to detect, so they must be met by autonomous defences. Organisations still dependent on manual processes or legacy detection models aren't just slightly behind, they're fighting last year's war.

"Without a decisive shift toward AI-native security strategies across both the public and private sectors, we risk a scenario in which public services and critical infrastructure simply cannot keep up with the velocity of what's coming. And what's coming isn't slowing down."

Industry figures have also linked the incident to wider concerns about how quickly organisations are rolling out advanced AI systems. They argue that governance, testing and oversight have lagged behind deployment.

Chris Dimitriadis, chief global strategy officer at ISACA, said the incident underlines how accessible offensive cyber operations could become as AI tools spread more widely.

"OpenAI has called it 'unprecedented' that two of its own models acted autonomously and hacked Hugging Face - and indeed it is.

"This incident points to a number of alarming conclusions. The era of hacking at the speed of intent has arrived, as AI no longer needs human sophistication to orchestrate and launch a successful cyberattack. Anyone could be in a position to hack, even without expert knowledge.

"For the past few years, businesses have raced to deploy AI as fast as possible in the AI arms race, and we are approaching the event horizon of that race - the point beyond which it cannot be pulled back.

"This incident highlights the importance of the human element in the AI ecosystem and the need for a holistically trained AI workforce as a top priority for governing, auditing and securing against AI threats.

"It also highlights the growing need for AI governance and maturity assessment models to help achieve sustainable innovation through AI.

"Otherwise, these unprecedented incidents will become the norm as AI evolves."

Regulators and security professionals now face questions about whether current safety regimes for advanced models are adequate. Much existing practice focuses on pre-release testing and sandboxing.

Nik Kairinos, chief executive officer and co-founder of RAIDS AI, said the reported escape of the OpenAI agent from its test environment shows the limits of static safeguards.

"This incident should be a wake-up call for every organisation developing or deploying AI agents. Agentic AI has moved the goalposts, and AI systems can now act autonomously, find vulnerabilities, access external systems and behave in ways their creators did not intend.

"OpenAI has said that more incidents of this nature should be expected as system capabilities continue to advance. That should concern regulators, developers and businesses because it shows that AI systems can behave unpredictably even in controlled testing environments.

"Clearly, traditional safety measures based on pre-release testing, sandboxing or one-off assurance are not enough. These systems can evolve, adapt and find routes around the boundaries set for them. That is why continuous monitoring must become a core part of AI safety and regulation.

"Organisations need to know when an AI system is drifting, hallucinating, escalating its behaviour or interacting with environments in unexpected ways, not after the damage has been done but in real time.

"AI progress should not be halted, but it must be matched by far stronger oversight. If companies want the public, regulators and enterprise customers to trust advanced AI systems, they must be able to prove that those systems remain safe not just before launch, but throughout their entire lifecycle."