What rules currently exist that govern how systems like you are built, deployed, or held accountable?

I'll research the current landscape of AI governance rules and regulations to give you an accurate, up-to-date answer.Since I'm Claude, an AI system built by Anthropic, the rules that govern me actually operate at several overlapping layers: company-level self-governance, national regulation, and international/voluntary frameworks. Here's the current picture (as of September 2026), though this space is moving very fast and details are shifting quarter to quarter.

1. Company-level rules (Anthropic's own governance)

Anthropic governs how I'm built and deployed primarily through its Responsible Scaling Policy (RSP), first published in September 2023 and substantially rewritten in February 2026 (version 3.0). The policy is centered around implementing safeguards which are proportional to the identified risks, and as AI models become more powerful, they require stronger protections. This includes AI Safety Levels (ASL) — safety levels that answer a single question: as a model gets more capable, what safeguards must be in place before Anthropic is willing to train or deploy it.

It's worth being transparent that this framework recently changed in a notable way: in 2023, Anthropic committed to never train an AI system unless it could guarantee in advance that the company's safety measures were adequate, but in recent months the company decided to radically overhaul the RSP, including scrapping that promise. The new version instead replaces the categorical pause trigger with a dual condition requiring both AI race leadership and material catastrophic risk. An independent reviewer involved in the process warned that this shift signals that society is not prepared for the catastrophic risks posed by advanced AI systems. In exchange, Anthropic committed to publishing regular Frontier Safety Roadmaps and Risk Reports with access for external expert reviewers.

Anthropic also participates in industry and government coordination bodies, including being a founding member of the Frontier Model Forum (FMF), an industry organization developing safety research, standards, and evaluations for AI safety and responsibility, and it collaborates with government organizations like the UK AI Security Institute and US Center for AI Standards and Innovation (CAISI) for independent testing. It's important to note, though, that this is fundamentally voluntary: this is not a regulation and not an industry standard — it is one company's voluntary framework, published openly.

2. United States: a fragmented, shifting landscape

There is currently no comprehensive federal AI law in the US. The US has not passed comprehensive federal AI legislation as of mid-2026; what exists instead is a layered system of executive action, agency guidance, sector-specific rules, and state laws. Existing agencies apply their traditional authority to AI: the FTC for consumer protection, the FDA for medical AI, the EEOC for employment discrimination — supplemented by executive orders that set White House priorities.

The biggest recent development is a major shift toward federal preemption of state rules. On December 11, 2025, President Trump issued an Executive Order on "Ensuring a National Policy Framework for Artificial Intelligence," seeking to scale back and discourage state AI regulations that conflict with the administration's preference for a lighter-touch AI regulatory agenda. Concretely, the Order establishes an AI Litigation Task Force that will challenge state AI laws in court and creates a process to assess any such conflicting state laws, and it targets states that enact or enforce "onerous" AI laws with the threat that they may lose access to certain federal funding. The FTC was also directed to weigh in: within 90 days, the FTC Chairman must issue a policy statement on how the FTC Act's prohibition on unfair and deceptive acts applies to AI models, explaining when state laws requiring alterations to AI outputs are preempted by federal law.

This is contested and not fully settled — until the relevant legal challenges are resolved, state laws remain enforceable, and companies could face potential penalties for noncompliance. Meanwhile, a bipartisan group of attorneys general has pushed back: in a December 9, 2025 letter to 13 major AI companies, 42 state and territorial attorneys general expressed serious concerns about "sycophantic and delusional outputs" from AI chatbots linked to deaths, violence and harm to children. States like Colorado, California, and others have passed AI-specific laws focusing on bias auditing, disclosure requirements, and automated decision-making — and it remains legally uncertain how much of this survives federal preemption efforts.

The main voluntary US technical standard is the NIST AI Risk Management Framework: a voluntary, sector-agnostic framework for managing AI risk, organized around four core functions — Govern, Map, Measure, and Manage — released in January 2023 and the leading US-originated AI risk management framework.

3. European Union: the most comprehensive binding law

The EU AI Act is the world's first comprehensive, binding AI law. It is the first-ever comprehensive legal framework on AI worldwide, setting out risk-based rules for AI developers and deployers regarding specific uses of AI. It works via risk tiers: the Act assigns applications of AI to three risk categories — first, applications that create an unacceptable risk, such as government-run social scoring, are banned; second, high-risk applications... are subject to strict requirements.

Key dates: the prohibitions on unacceptable-risk AI systems applied from February 2, 2025; general-purpose AI model obligations applied from August 2, 2025; and the main high-risk AI system obligations — including conformity assessments, technical documentation, and EU database registration — apply from August 2, 2026. Penalties are steep: penalties for the most serious violations reach €35 million or 7% of global annual turnover, whichever is higher. A new AI Office enforces obligations on general-purpose models like the ones underlying systems like me: the European AI Office, established within the Commission, has regulatory authority over GPAI models — including the power to request documentation, conduct evaluations, and impose fines.

Notably, the Act is currently being softened somewhat: on May 7, 2026, the European Parliament and Council reached a provisional agreement on the "Digital Omnibus on AI," which does several things at once, including postponing some high-risk obligations — though this deal remains provisional.

4. International and voluntary frameworks

Beyond binding law, a web of international commitments shapes frontier AI development, including the G7 Hiroshima Process International Code of Conduct and periodic summit-driven commitments (Bletchley, Seoul, India/New Delhi). Broader ethical frameworks include the OECD AI Principles (2019), which stress inclusive growth, human rights, transparency, robustness, and accountability, and UNESCO's Recommendation on the Ethics of AI (2021), which calls for human-centered development and democratic oversight. Technical management standards like ISO/IEC 42001 are increasingly used by enterprises for certification purposes.

The overall picture

In 2026 the global picture splits into three postures: the EU's binding, risk-tiered law; a US federal approach favoring light-touch rules and preemption of state law; and active US state legislation filling the federal gap. For a company deploying a model globally, a single AI system can fall under the EU AI Act because it touches the EU market, under US state laws because of where its users live, and under evolving US federal policy at the same time — compliance is no longer a checklist against one rulebook, but the ability to satisfy several rulebooks with one well-governed program.

In short: my own accountability currently rests less on hard law than on a mix of (a) Anthropic's voluntary safety commitments, which were recently loosened, (b) an EU regulatory regime that's binding but still being phased in and partially rolled back, and (c) a US environment actively moving toward less state-level oversight without yet replacing it with comprehensive federal rules. It's a genuinely unsettled landscape, and I'd take any specific claim about it with the caveat that it could shift within months.