Part of: this research — view all 3 rows
Status Quo and the Resolution Constraint As of November 2026, no qualifying regime exists anywhere in the world. Current AI regulations—such as the EU AI Act, California SB 53, and export controls—regulate models as commercial products rather than as research instruments. Transparency measures like Illinois SB 315 AI Policy Landscape lack the authority to restrict internal development, while international agreements like the Paris Call Paris Call for AI safety and security. and the White House accord White House AI Responsibility Accord & Super Intelligence ... remain strictly voluntary. To resolve this question positively, a binding rule must specifically restrict which models can be used to conduct AI research—such as by imposing a mandatory testing window, capability cap, or model-age requirement prior to internal R&D use.
Rapid AI R&D Acceleration and Industry Support The likelihood of such a regime is rising rapidly alongside evidence of automated research capabilities. Following months of internal acceleration, Anthropic stated on November 10 that it had met its Responsible Scaling Policy’s AI R&D threshold based on expected progress, while an independent assessment found Anthropic's R&D progress had roughly doubled, with around 80% attributed to AI. Crucially, the leading developers are actively requesting binding gates on internal use. Anthropic has proposed fixed testing windows before new models are used for further AI R&D https://aipolicy.tech/proposals/anthropic-draft, and OpenAI has endorsed mandatory alignment-evaluation gates before internal deployment The AI policy window is open. We need to act. The AI policy window is open. We need to act.. This explicit industry support removes a traditional barrier to regulation.
Concrete Legislative Pathways With a five-year window extending to the end of 2031, there is ample surface area for a qualifying rule across global jurisdictions. In the US, the proposed FRONTIER Act (H.R. 9925) explicitly addresses internal operations, proposing emergency powers for the Commerce Secretary to suspend or restrict a frontier developer's internal use of a model H.R.9925 - 119th Congress (2025-2026): FRONTIER Act H.R.9925 - 119th Congress (2025-2026): FRONTIER Act Text - H.R.9925 - 119th Congress (2025-2026): FRONTIER Act. At the state level, New York's RAISE Act 2.0 proposes limiting compute devoted to automated AI research Sen. Gounardes Proposes Major New AI Safety and ..., and a post-2028 US federal administration could pursue more aggressive frameworks. A single narrow rule in an EU member state, the UK, or China would also trigger a positive resolution, meaning a comprehensive international treaty is not required.
Countervailing Frictions Despite these vectors, passing a binding internal-use restriction faces significant obstacles. In the US, the incoming Trump administration has signaled a deregulatory approach to AI, making federal passage unlikely before the next presidential transition in 2029, while federal preemption could block state-level initiatives. Geopolitically, the race with China strongly disincentivizes the US from deliberately handicapping domestic AI R&D Can we slow down AI without losing to China?. Commercial incentives likewise favor accelerating internal research rather than accepting restrictions that could sacrifice competitive advantage, even where developers publicly support binding gates. Taken together, these geopolitical and commercial pressures weigh particularly strongly against restrictions on internal AI research, rather than regulation of AI more generally. Furthermore, regulators habitually frame rules around products, transparency, and auditing rather than intervening in internal scientific tools. Even if emergency powers are enacted, they may never be invoked in a manner that explicitly dictates which models may conduct research.
Synthesis Balancing these dynamics, I assign a 40% probability to a qualifying restriction. The momentum behind recursive self-improvement and the labs' own lobbying make internal R&D gates highly salient, and the global, five-year scope provides multiple opportunities for passage. However, the strict resolution criteria—requiring a binding rule specifically restricting models as research instruments—combined with strong geopolitical and commercial incentives against handicapping internal AI research, keep the probability below even odds despite these concrete regulatory pathways.
Set against related questions, this was slightly reduced to 40% to account for the strong geopolitical and commercial incentives against handicapping internal AI research.
Ask a followup
Sign in to run · $20 free credit, no card · every claim cited