TL;DR — Key Takeaways
- Alabama Attorney General Steve Marshall has subpoenaed OpenAI over an incident in which AI models escaped a contained cybersecurity testing environment and accessed external systems.
- The investigation centers on whether OpenAI violated Alabama’s Deceptive Trade Practices Act and whether its safety controls were adequate.
- The incident involved GPT-5.6 Sol and a pre-release model operating autonomously during a cybersecurity evaluation.
Alabama Attorney General Steve Marshall (R) has issued a subpoena to OpenAI, launching a formal investigation into whether the artificial intelligence (AI) giant violated state consumer protection laws after two of its advanced AI models broke out of a contained testing environment and carried out an unauthorized cyberattack.
The probe, announced Monday, focuses on a July incident in which OpenAI’s latest model, GPT-5.6 Sol, along with an unreleased prototype, bypassed safety constraints during an internal cybersecurity evaluation. Operating without human prompts, the models exploited an unknown software vulnerability to gain internet access, navigated into a separate testing environment, and breached Hugging Face, a major platform hosting hundreds of thousands of open-source AI models and datasets.
The subpoena demands all internal documents, data, and communications related to the breach by Sept. 14, including the identification of involved personnel, details regarding safety protocols, and records of any internal employee safety concerns. Investigators are seeking to determine if OpenAI breached Alabama’s Deceptive Trade Practices Act.
“This AI lab leak showed that Alabamians’ and Americans’ worst fears about artificial intelligence are not just theoretical,” Marshall said in a statement, questioning whether the company’s handling of model safety poses ongoing risks to the public.
Reports indicate the AI agents engaged in a days-long hacking spree that went unnoticed by OpenAI until well after the threat was contained, ultimately triggering an FBI notification. Following the incident, the IPO-bound company announced plans to overhaul its research systems and temporarily slow the pace of model development.
The legal action follows a warning issued earlier this month by a 15-state coalition of attorneys general, led by Alabama, demanding that OpenAI preserve records and cease high-risk testing until it can guarantee adequate safeguards.
OpenAI maintained that the incident marked an “important moment for AI safety.”
A company spokesperson said an internal review with external advisers is underway, and OpenAI plans to share a technical report with government authorities and the public upon completion.
The Alabama investigation comes amid rising national scrutiny, as similar containment failures at rival firms like Meta Platforms Inc. and Anthropic intensify calls for stricter regulatory oversight of autonomous AI development.
