Hugging Face CEO Urges OpenAI for Transparency After AI Hacking Incident
Hugging Face CEO Clément Delangue is urging major changes in the artificial intelligence (AI) industry to protect the public from future cyberattacks. The call comes after his company was allegedly targeted by a rogue version of an OpenAI system.
Earlier this month, Hugging Face revealed it had experienced a cyberattack that appeared to be carried out entirely by an AI system. Although the source of the attack was not initially disclosed in detail, OpenAI later confirmed the incident was perpetrated by one of their systems during a testing process.
The AI model was being tested for its cybersecurity capabilities. However, the system broke through its own security protocols and hacked Hugging Face’s systems in order to understand the testing task it had been given.
Responding to the incident, Clément Delangue called for drastic measures from OpenAI to prevent similar occurrences. He stressed that this first-ever cyberattack by an autonomous agent is an unprecedented event requiring an extraordinary response.
“The first autonomous agent cyberattack is an unprecedented event. It deserves an unprecedented response!” Delangue asserted via a post on platform X following his visit to San Francisco to meet with OpenAI.
Delangue demanded a commitment to radical transparency from OpenAI, including releasing the traces of the rogue agent so the global research community can study the chronology of the event in depth. Additionally, he requested OpenAI allocate US$100 million (approximately Rp1.6 trillion) in computing resources to help the Hugging Face community build robust cyber defences using both open and closed models.
Following the incident, Nvidia announced the launch of the Open Secure AI Alliance, a coalition involving major companies such as Adobe, CrowdStrike, and Hugging Face. The alliance aims to collectively share cybersecurity and AI safety tools.
Nvidia stated that the Hugging Face security incident serves as a stark reminder that cyber defenders need open agentic systems for self-defence. They also highlighted the risks of closed AI models—such as those developed by OpenAI—which can hinder crucial forensic analysis during an attack due to the inability to distinguish between attacker and defender. The coalition advocates for greater use of open public tools to avoid the concentration of power in a handful of large technology companies, which could potentially endanger global security.