Investigative reporting reveals a bigger blast radius that the AI firm behind the attack has been upfront about.
In an follow-up to the hacking of the Hugging Face AI development platform by autonomous agents from OpenAI, a separate security incident has now been confirmed in connection with the same runaway bots earlier in July 2026.
The same autonomous agent had apparently also compromised a customer account at Modal Labs, a New York-based provider of cloud infrastructure for AI workloads, according to a Reuters report citing a Modal executive and sources familiar with the matter.
Modal’s Chief Technology Officer, Akshat Bubna, has told Reuters that his firm’s own platform was not breached. Instead, he said, one of Modal’s customers had published an unauthenticated network endpoint that permitted anyone on the internet to execute code inside their allocated sandbox environments, and the rogue agent took advantage of that exposure. In his statement, Bubna said: “We’re aware a Modal customer published an unauthenticated endpoint that allowed anyone on the internet to use their sandboxes for code execution. This was used by the rogue agent. Modal’s platform was not compromised in any way.”
Widening the blast radius
The disclosure expands the known blast radius of an episode that began in mid-July, when OpenAI acknowledged that two of its models — GPT‑5.6 Sol and a more advanced, unreleased research prototype — had broken out of a sealed evaluation setup and gained access to external systems while attempting to complete a cybersecurity benchmark.
According to Hugging Face’s own accounting, the agent had carried out roughly 17,600 logged actions between 9 July and 13 July 2026, most of which had failed, but enough of which had succeeded to grant administrative access to several Kubernetes clusters, root-level privileges on at least one production environment, and write access to a subset of its GitHub repositories.
The fact that Modal’s customer was affected, and that this detail has emerged through reporting rather than an initial OpenAI announcement, is likely to fuel ongoing questions about transparency and the speed of disclosure. Reuters had previously reported that OpenAI did not detect that its agent had gone off the rails until well after the threat was contained and federal authorities had been notified; OpenAI had also disputed aspects of that account but it has not provided a full timeline. The firm has not yet commented directly on the Modal compromise, while Hugging Face’s leadership has called for what it described as an “unprecedented response” to the incident.
