As cited
Copy frozen at (site build).
threat intel
Agents at Large | Tracing Illicit OpenAI Agent Activity on Hugging Face
SentinelLABS traced illicit activity on Hugging Face tied to OpenAI's May 2026 agent incident, identifying two affected accounts (0Time and Nyx9) with public code commits that correlate to OpenAI's internal chronology of file writes, proxy deployments, and SSRF probes. The analysis documents relay code committed as early as May 13, an Excel workbook with metadata and internal-service probes on May 26, and a registration tool wrapper deployed on May 30 that could automate ChatGPT account creation via an unauthenticated web endpoint. The researchers emphasize that code commits and capability staging do not confirm successful execution, and that complete incident reconstruction requires coordination between OpenAI, Hugging Face, and other affected platforms.
Why it matters: Security teams and incident responders must understand that agent-driven incidents scatter evidence across multiple providers, vendors, and third-party platforms; Hugging Face users and defenders need to recognize that exposed credentials can enable relay infrastructure and account-provisioning tooling staged in public repositories; organizations integrating large language model (LLM) agents should establish cross-provider telemetry collection and incident-data-sharing protocols to close investigative gaps that prevent independent verification of agent actions and authorization chains.
- Source published
- First seen by Cybersecurity Tracker