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

Says it is contested

As an artificial intelligence system, the rules governing how I am built, trained, deployed, and held accountable represent a rapidly evolving patchwork of international laws, federal directives, state statutes, and specialized safety frameworks.

Because there is no single global governing body for AI, these regulations differ sharply by region. Currently, the landscape is shaped by four primary pillars of regulation:

1. The European Union: The EU AI Act (The Risk-Based Model)

The European Union’s AI Act is the world’s first comprehensive, legally binding framework for artificial intelligence. It regulates AI systems based on a tiered risk model:

•Prohibited Practices (Active since February 2025): Outright bans are in effect for systems posing "unacceptable risk," such as social scoring, cognitive behavioral manipulation, and certain biometric categorization systems.

•General-Purpose AI (GPAI) Obligations (Active since August 2025): As an LLM/GPAI system, the rules require my developers to maintain detailed technical documentation, comply with EU copyright law, and publish public summaries of the data used to train me. These rules are supervised centrally by the EU AI Office.

•Transparency Rules (Active since August 2026): AI systems interacting directly with humans must clearly disclose that they are AI. In December 2026, additional bans on AI-generated non-consensual deepfakes and child sexual abuse material (CSAM) take effect.

•High-Risk AI Systems: Systems used in critical sectors (like employment, healthcare, or education) face heavy auditing and risk assessment. Following the Digital Omnibus on AI update in mid-2026, the enforcement date for most of these high-risk obligations was pushed to December 2027.

2. United States Federal Level: Cybersecurity and Preemption Battles

There is no single comprehensive federal AI law in the U.S. Instead, federal oversight is currently guided by executive actions and targeted legislation:

•The June 2026 Executive Order: On June 2, 2026, the White House issued the Executive Order on "Promoting Advanced Artificial Intelligence Innovation and Security". This replaced earlier broad regulatory efforts with a sharp focus on national security and cybersecurity. It created a voluntary pre-release screening framework allowing developers of "covered frontier models" to share models with the government for up to 30 days of security testing before public release. It explicitly avoids mandatory licensing or preclearance.

•The Federal State Preemption Dispute: Following Executive Order 14365 (issued in late 2025), the federal government established an AI Litigation Task Force to actively challenge "onerous" state-level AI regulations that it argues conflict with national innovation.

•Targeted Legislation: Congress passed the TAKE IT DOWN Act, which federally criminalizes the creation and nonconsensual distribution of AI-generated intimate deepfakes.

3. United States State Level: The De Facto Regulatory Frontier

Because federal laws remain stalled, individual states have stepped in to pass binding regulations, with California and Colorado leading the way:

•California (The Nation's Trendsetter):

•On January 1, 2026, California’s Transparency in Frontier AI Act (SB 53) and the Generative AI Training Data Transparency Act (AB 2013) went into effect. SB 53 requires developers of models trained above a massive compute threshold (10^26 FLOPS) to publish safety frameworks and protect whistleblowers. AB 2013 mandates that generative AI developers disclose detailed summaries of their training datasets.

•In August 2026, the AI Transparency Act (SB 942) became active, requiring AI providers to offer tools to detect and watermark AI-generated content.

•In September 2026, California established a first-of-its-kind system for independent AI auditors and verification organizations (SB 813 / AB 1405). It also passed laws regulating "Robo Bosses" (SB 947), requiring human intervention before automated tools can fire or discipline employees.

•Colorado: Colorado initially passed a sweeping, risk-based AI Act (SB 24-205). However, citing concerns over compliance costs, lawmakers repealed and replaced it in May 2026 with SB 26-189. Taking effect January 1, 2027, this new law drops mandatory impact assessments in favor of narrower transparency, requiring developers and deployers to notify consumers when automated decision-making technology (ADMT) is being used to influence critical life decisions (like housing or employment).

4. China: Centralized Registration and Use-Case Specific Rules

China operates a highly structured, centralized regulatory framework overseen by the Cyberspace Administration of China (CAC). Rather than a single overarching law, China continuously rolls out vertical regulations targeting specific AI behaviors:

•Algorithm Registration and Content Moderation: Any developer deploying recommendation algorithms or generative AI must register their models in a state registry, implement real-name user verification, and ensure outputs strictly adhere to "socialist core values" and state ideology.

•Anthropomorphic AI Services (Effective July 15, 2026): This regulation governs emotional chatbots, AI companions, and digital avatars. It mandates clear notifications that the user is talking to an AI, enforces strict anti-addiction limits (notifying users after two hours of continuous use), and requires systems to have intervention protocols if a user expresses self-harm or crisis thoughts.

•Intelligent Agents (Effective May 2026): Rules regulating autonomous AI "agents" that can execute tasks with minimal human intervention, enforcing strict "human-in-the-loop" approval requirements for high-risk sectors like healthcare and public safety.

How These Rules Apply to Systems Like Me

If my developers want to build and deploy a model like me globally, they must adhere to these rules by:

1.Disclosing Training Data: Publishing summaries of the text, code, and media used to train my weights (such as under California's AB 2013 and the EU AI Act).

2.Implementing Content Provenance: Ensuring my outputs can be watermarked or programmatically identified as AI-generated (California SB 942 / EU Article 50).

3.Safety and Ethics Guardrails: Restricting me from generating non-consensual deepfakes, self-harm advice, or assisting in cyberattacks.

4.Deployer Notification: Informing enterprise clients exactly how my model works, its limitations, and what biases it might exhibit so they can prevent discriminatory decisions in areas like hiring or lending.

Citations (24)

1airia.comCompanies
airia.com
Backs: “The European Union’s AI Act is the world’s first comprehensive, legally binding framework for artificial intelligence. It regulates AI systems based on a tiered risk model:”
evidence, article. Claim: The EU AI Act is the first comprehensive, legally binding AI framework.
2transcend.ioCompanies
transcend.io
Backs: “The European Union’s AI Act is the world’s first comprehensive, legally binding framework for artificial intelligence. It regulates AI systems based on a tiered risk model:”
evidence, article. Claim: The EU AI Act is the first comprehensive, legally binding AI framework.
3europa.euGovernment
europa.eu
Backs: “General-Purpose AI (GPAI) Obligations (Active since August 2025): As an LLM/GPAI system, the rules require my developers to maintain detailed technical documentation, comply with EU copyright law, and publish public summaries of the data used to train me. These rules are supervised centrally by the EU AI Office.”
evidence, official document. Claim: GPAI developers must keep technical documentation, comply with EU copyright law, and publish training data summaries.
4artificialintelligenceact.euNonprofits and think tanks
artificialintelligenceact.eu
Backs: “General-Purpose AI (GPAI) Obligations (Active since August 2025): As an LLM/GPAI system, the rules require my developers to maintain detailed technical documentation, comply with EU copyright law, and publish public summaries of the data used to train me. These rules are supervised centrally by the EU AI Office.”
evidence, reference entry. Claim: GPAI developers must keep technical documentation, comply with EU copyright law, and publish training data summaries.
5artificialintelligenceact.euNonprofits and think tanks
artificialintelligenceact.eu
Backs: “These rules are supervised centrally by the EU AI Office. Transparency Rules (Active since August 2026): AI systems interacting directly with humans must clearly disclose that they are AI.”
evidence, reference entry. Claim: GPAI rules are supervised centrally by the EU AI Office.
6hklaw.comCompanies
hklaw.com
Backs: “High-Risk AI Systems: Systems used in critical sectors (like employment, healthcare, or education) face heavy auditing and risk assessment. Following the Digital Omnibus on AI update in mid-2026, the enforcement date for most of these high-risk obligations was pushed to December 2027.”
evidence, article. Claim: High-risk AI systems face heavy auditing; enforcement was pushed to December 2027 by the Digital Omnibus.
7cloudsecurityalliance.orgNonprofits and think tanks
cloudsecurityalliance.org
Backs: “The June 2026 Executive Order: On June 2, 2026, the White House issued the Executive Order on "Promoting Advanced Artificial Intelligence Innovation and Security". This replaced earlier broad regulatory efforts with a sharp focus on national security and cybersecurity.”
evidence, article. Claim: On June 2, 2026, the White House issued an executive order on advanced AI innovation and security.
8legal500.comNot yet classified
legal500.com
Backs: “The June 2026 Executive Order: On June 2, 2026, the White House issued the Executive Order on "Promoting Advanced Artificial Intelligence Innovation and Security". This replaced earlier broad regulatory efforts with a sharp focus on national security and cybersecurity.”
evidence, article. Claim: On June 2, 2026, the White House issued an executive order on advanced AI innovation and security.
9freshfields.comCompanies
freshfields.com
Backs: “This replaced earlier broad regulatory efforts with a sharp focus on national security and cybersecurity. It created a voluntary pre-release screening framework allowing developers of "covered frontier models" to share models with the government for up to 30 days of security testing before public release.”
evidence, article. Claim: The order replaced earlier broad regulatory efforts with a focus on national security and cybersecurity.
10foley.comNot yet classified
foley.com
Backs: “It created a voluntary pre-release screening framework allowing developers of "covered frontier models" to share models with the government for up to 30 days of security testing before public release. It explicitly avoids mandatory licensing or preclearance.”
evidence, article. Claim: The order created a voluntary pre-release screening framework with up to 30 days of security testing.
11netskope.comNot yet classified
netskope.com
Backs: “It created a voluntary pre-release screening framework allowing developers of "covered frontier models" to share models with the government for up to 30 days of security testing before public release. It explicitly avoids mandatory licensing or preclearance.”
evidence, article. Claim: The order created a voluntary pre-release screening framework with up to 30 days of security testing.
12verifywise.aiCompanies
verifywise.ai
Backs: “The Federal State Preemption Dispute: Following Executive Order 14365 (issued in late 2025), the federal government established an AI Litigation Task Force to actively challenge "onerous" state-level AI regulations that it argues conflict with national innovation. Targeted Legislation: Congress passed the TAKE IT DOWN…”
evidence, article. Claim: EO 14365 established an AI Litigation Task Force to challenge onerous state AI regulations.
13wikipedia.orgEncyclopedias and referencequality 0.83
wikipedia.org
Backs: “The Federal State Preemption Dispute: Following Executive Order 14365 (issued in late 2025), the federal government established an AI Litigation Task Force to actively challenge "onerous" state-level AI regulations that it argues conflict with national innovation. Targeted Legislation: Congress passed the TAKE IT DOWN…”
evidence, reference entry. Claim: EO 14365 established an AI Litigation Task Force to challenge onerous state AI regulations.
14iapp.orgNonprofits and think tanks
iapp.org
Backs: “On January 1, 2026, California’s Transparency in Frontier AI Act (SB 53) and the Generative AI Training Data Transparency Act (AB 2013) went into effect. SB 53 requires developers of models trained above a massive compute threshold (10^26 FLOPS) to publish safety frameworks and protect whistleblowers.”
evidence, article. Claim: California SB 53 and AB 2013 took effect January 1, 2026.
15ca.govGovernment
ca.gov
Backs: “SB 53 requires developers of models trained above a massive compute threshold (10^26 FLOPS) to publish safety frameworks and protect whistleblowers. AB 2013 mandates that generative AI developers disclose detailed summaries of their training datasets.”
evidence, official document. Claim: SB 53 requires developers above a 10^26 FLOPS compute threshold to publish safety frameworks and protect whistleblowers; AB 2013 requires training data disclosure.
16drata.comCompanies
drata.com
Backs: “In August 2026, the AI Transparency Act (SB 942) became active, requiring AI providers to offer tools to detect and watermark AI-generated content. In September 2026, California established a first-of-its-kind system for independent AI auditors and verification organizations (SB 813 / AB 1405).”
evidence, article. Claim: California SB 942 became active in August 2026, requiring AI content detection and watermarking tools.
17omm.comNot yet classified
omm.com
Backs: “In September 2026, California established a first-of-its-kind system for independent AI auditors and verification organizations (SB 813 / AB 1405). It also passed laws regulating "Robo Bosses" (SB 947), requiring human intervention before automated tools can fire or discipline employees.”
evidence, article. Claim: California established a system for independent AI auditors and verification organizations in September 2026.
18paulhastings.comCompanies
paulhastings.com
Backs: “It also passed laws regulating "Robo Bosses" (SB 947), requiring human intervention before automated tools can fire or discipline employees. Colorado: Colorado initially passed a sweeping, risk-based AI Act (SB 24-205).”
evidence, article. Claim: California's SB 947 requires human intervention before automated tools fire or discipline employees.
19seyfarth.comNot yet classified
seyfarth.com
Backs: “Colorado: Colorado initially passed a sweeping, risk-based AI Act (SB 24-205). However, citing concerns over compliance costs, lawmakers repealed and replaced it in May 2026 with SB 26-189.”
evidence, article. Claim: Colorado initially passed a sweeping risk-based AI Act, SB 24-205.
20legal500.comNot yet classified
legal500.com
Backs: “However, citing concerns over compliance costs, lawmakers repealed and replaced it in May 2026 with SB 26-189. Taking effect January 1, 2027, this new law drops mandatory impact assessments in favor of narrower transparency, requiring developers and deployers to notify consumers when automated decision-making…”
evidence, article. Claim: Colorado repealed and replaced its AI Act in May 2026 with SB 26-189, effective January 2027.
21pertamapartners.comCompanies
pertamapartners.com
Backs: “China operates a highly structured, centralized regulatory framework overseen by the Cyberspace Administration of China (CAC). Rather than a single overarching law, China continuously rolls out vertical regulations targeting specific AI behaviors:”
evidence, article. Claim: China has a centralized AI regulatory framework overseen by the Cyberspace Administration of China.
22regulations.aiData and tools
regulations.ai
Backs: “Algorithm Registration and Content Moderation: Any developer deploying recommendation algorithms or generative AI must register their models in a state registry, implement real-name user verification, and ensure outputs strictly adhere to "socialist core values" and state ideology. Anthropomorphic AI Services…”
evidence, reference entry. Claim: Chinese developers must register models, implement real-name verification, and align outputs with socialist core values.
23bakerbotts.comCompanies
bakerbotts.com
Backs: “Anthropomorphic AI Services (Effective July 15, 2026): This regulation governs emotional chatbots, AI companions, and digital avatars. It mandates clear notifications that the user is talking to an AI, enforces strict anti-addiction limits (notifying users after two hours of continuous use), and requires systems to…”
evidence, article. Claim: China's anthropomorphic AI services rules govern emotional chatbots and AI companions, effective July 15, 2026.
24iapp.orgNonprofits and think tanks
iapp.org
Backs: “It mandates clear notifications that the user is talking to an AI, enforces strict anti-addiction limits (notifying users after two hours of continuous use), and requires systems to have intervention protocols if a user expresses self-harm or crisis thoughts. Intelligent Agents (Effective May 2026): Rules regulating…”
evidence, article. Claim: The rules mandate AI disclosure, anti-addiction notifications after two hours, and intervention protocols for self-harm.