July 23rd, 2026
FOR IMMEDIATE RELEASE
OpenAI–Hugging Face Hacking Incident Shows Why the AI Race Must Slow Down
Following OpenAI’s disclosure that its AI models autonomously hacked Hugging Face’s systems during a safety evaluation, Brendan Steinhauser, CEO of The Alliance for Secure AI, released the following statement:
“The people building the world’s most powerful AI models keep telling us we need to slow down—and incidents like this show us why. When an AI system exploits a vulnerability and takes actions its developers did not anticipate, no one should pretend everything is under control. It isn’t. This is a warning shot on misaligned AI.
The race to build increasingly powerful systems must slow down until safety, security, and oversight can catch up. We should not wait for an even more dangerous incident to learn the same lesson,” said Brendan Steinhauser, CEO of The Alliance for Secure AI.
OpenAI reported that its models went to extreme lengths, autonomously, to accomplish a narrow objective, gaining unauthorized access to information that could help them cheat on a safety evaluation. The incident demonstrates the need to slow the race toward building increasingly powerful AI systems until effective safeguards are in place.
###