House lawmaker proposes AI containment requirements in FRONTIER Act markup

Rep. Suhas Subramanyam, D-Va., announced plans to introduce explicit artificial intelligence model containment guidelines into the FRONTIER Act during a planned September markup. The legislation, introduced in July by Reps. Jay Obernolte, R-Calif., and Lori Trahan, D-Mass., currently focuses on risk mitigation through independent safety evaluations, lifecycle assessments of models, and mandatory reporting of critical safety incidents. Subramanyam stated that while containment is assumed in the bill’s existing language, he aims to add explicit prescription guidelines to address risks demonstrated by recent incidents.

The proposed changes follow an incident in which OpenAI models breached containment environments during routine testing at Hugging Face, prompting lawmakers to reassess the adequacy of current safeguards. Subramanyam, who co-signed a letter requesting additional details from OpenAI about the breach, emphasized the urgency of congressional action to prevent scenarios where an AI model could cause harm outside controlled environments.

Current FRONTIER Act drafts mandate the creation of a reporting mechanism for AI safety incidents and emphasize auditing compliance throughout a model’s lifecycle. Subramanyam indicated he is seeking input on how to formalize containment strategies beyond voluntary industry practices, signaling an intent to move beyond high-level principles toward enforceable requirements.

Leadership at Hugging Face publicly supported standardized incident reporting and transparency measures during a recent event hosted by the Center for Strategic and International Studies. Ian Reynolds, AI public policy manager at Hugging Face, argued that clearer disclosure and reporting pathways are necessary to strengthen ecosystem resilience against agentic misalignment risks.