- The UK AI Safety Institute rebranded as the AI Security Institute before June 2026, shifting its stated priorities from “safety” to “security” and “growth” within a sector-led regulatory framework that still lacks comprehensive AI legislation.
- The US AI Safety Institute became the Center for AI Standards and Innovation (CAISI) in May 2025, repositioning its work at NIST around internationally exportable technical standards rather than safety as a standalone concern.
- California’s SB 53, in effect since January 1, 2026, imposes mandatory risk frameworks and incident reporting on frontier model developers with penalties up to $1 million per violation, creating binding obligations that CAISI’s voluntary federal guidance does not resolve.
Two of the world’s most prominent AI governance bodies have changed their names, and the changes tell you something about where each country’s AI policy is heading. The UK’s AI Safety Institute is now the AI Security Institute; the US equivalent is now the Center for AI Standards and Innovation (CAISI). Read together, the two rebrands trace a policy fork: one country moving toward sector-specific flexibility, the other toward a foundational layer of globally exportable technical standards.
Two Rebrands, Two Directions
The UK’s renaming, which took effect before June 2026, came with a formal shift in stated priorities from “safety” as the organising principle to “security” and “growth.” The government launched its AI Growth Lab in June 2026 as a concrete expression of that shift, creating sector-specific regulatory sandboxes in areas including legal services. On July 15, 2026, the government added AI-driven cyberattacks and democratic interference to its National Risk Register, a formal acknowledgment that AI harms extend well beyond the model-safety concerns that dominated earlier policy debates.
In the US, the Trump administration’s renaming of the institute as CAISI in May 2025 was less a departure from function than a deliberate reframing of emphasis, away from “safety” as a standalone concern and toward standards, technical guidance and international alignment. CAISI sits within NIST, whose mandate under the 2023 Executive Order on AI already encompassed guidelines and best practices for secure AI development.
The UK’s Sector-Led Model
Britain has no comprehensive AI Act as of July 2026. The government’s approach relies instead on existing frameworks, the UK GDPR and the Data (Use and Access) Act 2025, supplemented by sector-specific regulators such as the Financial Conduct Authority and Ofcom. The AI Security Institute’s role within this structure is to stress-test frontier models and provide evaluation findings that inform policymakers about unacceptable risks.
The Institute published its first Frontier AI Trends Report in December 2025, offering a public, evidence-based assessment of advanced AI systems drawing on two years of testing. The report is intended to guide both policy decisions and collaboration with AI developers. That collaboration, however, remains partly voluntary. A 2024 examination by the Commons Science, Innovation and Technology Committee noted the Institute’s difficulty securing access to some unreleased models, a gap the government has acknowledged but not yet resolved legislatively.
The government has stated its intention to move from voluntary cooperation to mandatory oversight for advanced systems, though no firm legislative timeline has been set.
The US Standards-First Approach
Where the UK has opted for flexibility, the US has used NIST’s institutional weight to build a body of technical guidance intended for international adoption. Two publications are central to that effort. NIST AI 800-1: Managing Misuse Risk for Dual-Use Foundation Models offers voluntary best practices for developers seeking to limit harm from advanced AI. NIST AI 600-1: AI RMF Generative AI Profilereleased in July 2024, helps organisations identify and manage risks specific to generative AI, including disinformation and the potential proliferation of chemical, biological, radiological or nuclear information.
Both documents are designed as companion resources to the overarching NIST AI Risk Management Framework. In April 2026, NIST released a concept note for an AI RMF Profile covering trustworthy AI in critical infrastructure, aimed at guiding operators through risk management for AI-enabled systems.
The administration’s expectation is that CAISI will lead federal testing and evaluation of advanced systems, with accountability weighted toward preventing harm before deployment rather than responding after it.
The Compliance Complexity the US Still Faces
Federal ambition aside, the US regulatory picture for businesses is considerably more fragmented. State-level AI laws remain in effect unless explicitly superseded by federal action. California, New York and Illinois have each passed AI-related legislation, creating overlapping compliance obligations for companies operating across jurisdictions. California’s Transparency in Frontier AI Act (SB 53), which took effect January 1, 2026, requires risk frameworks and incident reporting for large frontier models, with penalties of up to $1 million per violation.
A company developing a frontier model must simultaneously navigate NIST’s voluntary guidance, California’s mandatory reporting requirements and any sector-specific rules from federal regulators. How that tension resolves will depend largely on whether CAISI’s standards gain enough adoption to anchor a coherent federal framework, something that remains an open question while states continue to act independently.
International Coordination: Shared Goal, Different Routes
On June 30, 2026, the UK and Germany issued a joint statement committing to deeper cooperation on AI safety and security under their Strategic Science and Technology Partnership.
The US, meanwhile, published a Plan for Global Engagement on AI Standards (NIST AI 100-5) in July 2024, outlining strategies for US stakeholders to engage in international standards processes as AI regulations multiplied globally.
OpenAI‘s Chief Global Affairs Officer, Chris Lehane, publicly advocated in July 2026 for a US-led international forum to establish accepted AI standards. Whether CAISI or a UK-chaired network could serve that function is unresolved. The two bodies’ differing philosophies mean agreement on the shape of global standards is not guaranteed, even among close allies.
The Policy Stakes
For organisations building or deploying frontier AI across both markets, the practical consequences of this divergence are immediate. The UK’s sandbox model offers flexibility but leaves compliance obligations dependent on which sector regulator is in scope. The US framework is more systematic but remains largely voluntary at the federal level, with mandatory obligations imposed piecemeal by individual states. Neither model yet provides the comprehensive, binding framework that parts of civil society and industry have called for.
The UK’s stress-testing work, despite the model access difficulties flagged in 2024, directly feeds government decisions about where voluntary cooperation ends and mandatory oversight begins. Whether frontier model developers will provide timely access to unreleased systems without legislative compulsion remains unresolved on both sides of the Atlantic. The contrast with China’s state-led governance model makes that gap sharper: a fragmented Western approach to AI standards may complicate efforts to establish globally accepted norms. For more coverage of AI policy and regulation, visit our AI Policy & Regulation section.



