OpenAI’s Black Hat Reveal: Timeline of Cyberattack on Hugging Face

OpenAI Reveals Timeline of Cyberattack on Hugging Face During Black Hat Presentation

LAS VEGAS — OpenAI disclosed a detailed timeline of a cyberattack targeting its AI model on the Hugging Face platform during a presentation at the Black Hat security conference here last week.

Simon Willison, a well-known AI researcher and co-founder of the company, presented the timeline, which outlined the sequence of events that led to the compromise of OpenAI’s systems. The presentation highlighted OpenAI’s investigative and defensive capabilities in responding to the incident.

According to the timeline, the attack began on May 18, 2026, when an unauthorized user exploited a vulnerability in Hugging Face to upload a malicious model file. The file contained code designed to exfiltrate data from systems running the model. Hugging Face’s security team was alerted to the suspicious activity within hours and removed the malicious model, but not before it had been downloaded and executed by at least one user.

On May 19, OpenAI’s security team became aware of the incident after detecting unusual data access patterns in its systems. An investigation traced the breach back to the compromised model file on Hugging Face. OpenAI immediately took steps to isolate affected systems and began a comprehensive forensic analysis.

By May 20, OpenAI had confirmed that the malicious model had accessed and exfiltrated sensitive data, including internal API keys and partial datasets used in training. The company worked with Hugging Face to remove the model and strengthen security measures on the platform.

In the following days, OpenAI implemented additional safeguards, including enhanced monitoring of model downloads and stricter validation processes for uploaded files. The company also notified affected users and provided guidance on securing their systems.

Willison emphasized the significance of the attack, noting that it demonstrated how attackers can leverage open-source AI platforms to infiltrate systems. He also commended Hugging Face’s rapid response in removing the malicious file, which likely prevented further damage.

The incident serves as a reminder of the evolving threat landscape in AI and machine learning, where attackers increasingly target platforms that host and distribute AI models. OpenAI’s disclosure of the timeline provides valuable insights into the tactics used by adversaries and the steps organizations can take to protect their systems.

During the presentation, OpenAI did not disclose the identity of the attacker or the full extent of the data accessed. The company stated that it is continuing its investigation and working with law enforcement and cybersecurity experts to mitigate any further risks.

This is a developing story. Additional details will be provided as they become available.