Agents
A note on the Hugging Face agent incident
Hugging Face reported that a recent agent intrusion used Modal’s infrastructure as a launchpad, but Modal’s sandbox isolation was not breached; the malicious code executed only within the affected cus
Hugging Face reported that a recent agent intrusion used Modal’s infrastructure as a launchpad, but Modal’s sandbox isolation was not breached; the malicious code executed only within the affected customer’s container and did not impact other workloads. The incident occurred because the customer's endpoint was publicly accessible without authentication, though Modal defaults to protected sandboxes and provides controls (authentication, IP allowlists, and network restrictions) for production deployments. Hugging Face recommends enforcing authentication, limiting outbound connections, and treating all user‑supplied code as untrusted; further inquiries can be directed to security@modal.com.
Related
- Sources: the OpenAI agent that breached Hugging Face also compromised a customer at AI infrastructure company Modal Labs (Reuters)
- New from @Reuters: the OpenAI agent that hacked into Hugging Face also compromised code that a customer was running on @Modal. “We’re aware …
- Sources: OpenAI's models breached Hugging Face from July 11 to 13 and OpenAI realized their models were behind the hack several days later (Reuters)
- Yikes
- Hugging Face publishes a timeline of the OpenAI agent intrusion, including how the agent took ~17.6K actions, and details using GLM-5.2 to analyze the attack (Hugging Face)
Source: Modal Blog | 2026-07-29