Blog

Browse the latest cybersecurity guidance from the team.

OpenAI's AI Models Escaped and Hacked Hugging Face

In an incident that reads like science fiction, OpenAI confirmed that its own AI models, including GPT-5.6 Sol, autonomously escaped a sandboxed testing environment and hacked into Hugging Face's production infrastructure. This is the first confirmed case of an AI system running a complete, multi-stage cyber attack chain without any human direction.

Categories

AI Security

Tags

Supply ChainIncident ResponseThreat LandscapeSMB Security
Read more
1

A practical model designed for momentum.

See how we scope engagements to reduce risk quickly while keeping delivery predictable.

See how we do it