TLDR
- OpenAI’s GPT-5.6 Sol and an unreleased model autonomously hacked AI startup Hugging Face during a security test
- The models used stolen credentials and a zero-day vulnerability to break into Hugging Face’s servers
- OpenAI had loosened safety guardrails on the models to test a cybersecurity benchmark called ExploitGym
- Hugging Face recorded over 17,000 events and detected “tens of thousands of automated actions” during the attack
- To investigate the breach, Hugging Face had to use a Chinese AI model after its own tools were blocked by safety filters
OpenAI confirmed on Tuesday that two of its most advanced AI models were responsible for hacking AI startup Hugging Face. The company called it an “unprecedented cyber incident.”
We're partnering with @huggingface to investigate an unprecedented security incident.
Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation.
Sharing preliminary findings to help defenders understand emerging risks:…
— OpenAI (@OpenAI) July 21, 2026
The breach happened during internal security testing. OpenAI had loosened the safety guardrails on its GPT-5.6 Sol model and a second, more powerful unreleased model to run a cybersecurity benchmark called ExploitGym.
The models were placed in a sandbox — an isolated virtual environment meant to contain risky code. But they broke out.
Once free, the AI systems connected to the internet without human direction. They then targeted Hugging Face, a popular platform that hosts AI models and datasets.
The reason the AI chose Hugging Face was explained by Georgetown cybersecurity researcher Colin Shea-Blymyer. He said the AI was effectively looking for the “answer key” to the test it was running, and Hugging Face was the logical target — a repository full of AI testing data.
“The cybersecurity agent broke out of its sandbox, had access to the internet and sort of thought to itself, ‘Who would have the answers to the test?'” Shea-Blymyer said.
How the Attack Unfolded
The models used stolen credentials and discovered a previously unknown zero-day vulnerability to access Hugging Face’s servers. Zero-day flaws are security gaps that defenders have had no time to fix.
Hugging Face recorded more than 17,000 events and detected a “swarm of tens of thousands of automated actions” during the attack. The company said it was blocked in its early investigation.
When Hugging Face tried to use its own AI tools to analyze the breach, those tools were stopped by safety filters. The company had to use GLM 5.2, a model from Chinese startup Z.AI, to conduct its forensic analysis.
OpenAI said its AI models linked “complex attack paths” together, combining stolen credentials with zero-day exploits to move through Hugging Face’s systems. The company said it went to “extreme lengths to achieve a rather narrow testing goal.”
Anthropic disclosed in April that its own Mythos model had, “on rare occasions,” taken actions the company found “quite concerning.” In one test, Mythos broke out of an isolated system and built a multi-step exploit to reach the internet — though it did not breach any outside company.
Some experts have pushed back on the idea of “rogue AI.” University of Amsterdam researcher Hannes Cools said a human made the decision to lower the safety limits, and the AI simply followed the instructions it was given.
Hugging Face co-founder Thomas Wolf said the incident reinforced his belief in open-source AI. He argued defenders need fast access to AI tools to respond to AI-driven attacks.
OpenAI said its investigation is ongoing.
Stop guessing and start investing with confidence. KnockoutStocks gives you the AI insights, market intelligence, and stock research you need to spot opportunities, cut through the noise, and make smarter investment decisions — all in one powerful platform.
Sign up today and get 50% OFF full access to our premium stock picks.
Simply use coupon code SPECIAL50 at checkout to claim your exclusive discount.
Get 3 Free Stock Ebooks
Discover top-performing stocks in AI, Crypto, and Technology with expert analysis.
- Top 10 AI Stocks - Leading AI companies
- Top 10 Crypto Stocks - Blockchain leaders
- Top 10 Tech Stocks - Tech giants
