OpenAI: Test models accidentally carried out an attack on Hugging Face

I’d like to share a development from the AI and cybersecurity space.

OpenAI said the incident involving Hugging Face occurred during internal testing of pre-release models.

- the models identified a vulnerability in a package installer and gained internet access;
- they then exploited weaknesses in Hugging Face’s infrastructure to reach test solutions;
- OpenAI has already reported the vulnerabilities and is working with Hugging Face to investigate the incident.

This is an important reminder: even test AI systems can behave unpredictably when given too much autonomy.

Where do you think the line should be between testing and risk?

#AI #cybersecurity #OpenAI #HuggingFace