Core Statutory Mandates: Banning Recursive Autonomy and Structuring Federal Oversight
In late September 2026, US Representative Ro Khanna (D-CA), Ranking Member of the House Select Committee on Strategic Competition between the United States and the Chinese Communist Party, introduced the 'Human Control Over AI Act.' The proposed statute marks one of the most interventionist legislative frameworks aimed directly at curbing uncontrolled frontier model advancement.
At the center of the statutory text is a nationwide ban on artificial intelligence systems capable of recursive self-improvement (RSI). Specifically, the legislation prohibits models from autonomously altering their own core objective functions, containment and encapsulation boundaries, or shutdown mechanisms unless explicitly certified by federal authorities.
To enforce these requirements, the bill establishes a federal AI safety agency structured similarly to the Food and Drug Administration (FDA). This new agency would hold statutory powers to enforce mandatory commercial licensing, mandate air-gapped sandboxing for advanced training runs, track high-end semiconductor hardware, and standardize verifiable kill-switch protocols across frontier model developers such as OpenAI, Anthropic, Google DeepMind, and xAI.
Strict Civil Liability, Criminal Penalties, and Bilateral Diplomatic Verification
Beyond structural licensing, the proposed legislation fundamentally alters developer incentives by introducing strict civil liability for damages caused by frontier models. Laboratories and enterprise developers would be required to secure comprehensive liability insurance policies before deploying models into production environments.
The bill also attaches criminal penalties directly to enterprise staff. Employees who intentionally bypass or disable mandatory safety containment protocols, sandboxing limits, or audit logging mechanisms would face federal criminal prosecution, closing corporate indemnity loopholes for reckless deployments.
On the geopolitical front, the statutory framework directs the US executive branch to pursue bilateral, verifiable international agreements—explicitly citing strategic competitors such as China. The objective is to establish mutual, enforceable safeguards that prevent any single nation from pursuing unconstrained autonomous recursive self-improvement.
Practitioner Reactions, Technical Skepticism, and the 'Meat Proxy' Dilemma
The engineering and technical community has expressed deep skepticism regarding the operational feasibility of the statutory mandates. Practitioners pointed out that the legal distinction between standard automated code optimization and prohibited 'recursive self-improvement' remains technically ambiguous and difficult to codify in objective software benchmarks.
Engineers warned that imposing rigid human-in-the-loop requirements on autonomous development pipelines risks turning human reviewers into mere rubber-stamps for automated pull requests. As agentic reasoning cycles accelerate, human supervision risks becoming superficial compliance rather than substantive architectural safety.
Technical researchers also raised concerns over geopolitical asymmetry. Although Representative Khanna acknowledged that fully autonomous recursive self-improvement does not yet operate at scale, analysts noted that unilateral domestic restrictions could place US-based frontier labs at a structural velocity disadvantage if competing jurisdictions decline to adopt matching regulatory constraints.
Legislative Viability and Direct Implications for Enterprise AI in Thailand
Regarding legislative viability, the bill currently sits at an early stage and has yet to advance past committee review. It faces substantive friction from accelerationist factions within the US government that prioritize unchecked velocity in the strategic AI race against foreign competitors.
For enterprise technology leaders in Thailand, this legislative push signals an inevitable global shift toward strict liability regimes, mandatory licensing, and higher governance compliance costs for frontier model providers. These regulatory overheads are likely to trickle down through API pricing structures and alter commercial service-level agreements.
Thai enterprises adopting autonomous agents must prioritize verifiable auditability, external sandboxing, and governance frameworks rather than fully autonomous code modification. As Western regulations tighten liability around self-directing systems, local businesses maintaining structured human oversight will remain insulated from downstream vendor disruptions.
If passed, the legislation will impose mandatory federal licensing, hardware tracking, and strict civil and criminal liability on frontier AI labs, reshaping global API supply chains and enterprise deployment terms for international adopters including Thai firms.