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

The regulatory landscape governing how large language models and generative AI systems are built, deployed, and held accountable is rapidly evolving. Today, a combination of sweeping global frameworks, targeted state-level laws, and administrative crackdowns dictate what AI developers can and cannot do.

These rules can be broken down into four major areas: global foundational regulations, U.S. state and federal frameworks, China's targeted administrative measures, and broader legal accountability mechanisms like copyright and privacy.

1. The Global Gold Standard: The EU AI Act

The European Union’s AI Act is the most comprehensive, binding, and globally influential piece of AI legislation. It regulates AI based on its potential to cause harm, with severe penalties of up to €35 million or 7% of global annual turnover for non-compliance.

•General-Purpose AI (GPAI) Rules: Large language models and foundation models are categorized as "GPAI models". Under rules that took effect in August 2025, developers must maintain rigorous technical documentation, comply with EU copyright law, publish detailed summaries of their training data, and share information with downstream developers who build apps on their models. Highly powerful models that pose "systemic risk" face even tighter safety testing and reporting mandates.

•Strict Transparency (Enforced August 2, 2026): The implementation of Article 50 mandates that users must be clearly informed when they are interacting with an AI (like a chatbot). Furthermore, providers must use invisible, machine-readable watermarks to label synthetic text, images, and video as AI-generated.

•Prohibited AI Practices: The EU strictly bans AI uses deemed to carry "unacceptable risk". This includes social scoring, biometric categorization based on sensitive characteristics, and emotion recognition in workplaces.

2. United States: The State-Level Push vs. Federal Deregulation

The United States currently has no comprehensive federal AI law equivalent to the EU AI Act. Instead, regulation is split between a "patchwork" of strict state-level laws and a federal push focused on global competitiveness and deregulation.

State-Level Regulations (The Primary Driver of Binding Rules)

Because Congress has not passed overarching AI safety legislation, states have stepped in:

•California’s Frontier AI Rules (In effect Jan 1, 2026): Under the Transparency in Frontier AI Act (SB 53), developers of "frontier models" (trained using more than $10^{26}$ FLOPs of compute) must publicly disclose their safety frameworks, explain how they mitigate "catastrophic risks," and implement strict whistleblower protections for employees.

•California's Training Data Disclosures (In effect Jan 1, 2026): AB 2013 (The Training Data Transparency Act) requires generative AI developers to publish high-level summaries on their websites detailing the data used to train, fine-tune, or test their models—specifically disclosing the use of personal data, copyrighted materials, and dataset sources.

•California's Watermarking Mandates (In effect Aug 2, 2026): Under SB 942 (the California AI Transparency Act), public generative AI services with over 1 million monthly users must offer free AI detection tools and embed latent provenance metadata inside generated media.

•Colorado & Texas: Texas’s TRAIGA (HB 149) went into effect on January 1, 2026, banning manipulative or harmful uses of AI (like government social scoring). Meanwhile, Colorado's heavily debated SB 24-205 was replaced in mid-2026 by SB 189, a streamlined law taking effect on January 1, 2027, that focuses on transparency and consumer disclosures regarding automated decision-making.

Federal Level: Deregulation and National Preemption

The federal posture, shaped by a series of executive orders, prioritizes stripping away regulatory red tape to accelerate domestic AI development:

•Executive Order 14179 (January 2025): Revoked the Biden-era AI safety executive order (EO 14110) in favor of removing barriers to AI adoption.

•Executive Order 14365 (December 2025): Directed federal agencies to challenge state-level AI laws (like those in California) through legal action if they are deemed too "onerous" or in conflict with national AI goals.

•Advanced AI Security EO (June 2, 2026): Created a voluntary testing framework allowing the federal government to run classified security and cyber-capability benchmarking on "covered frontier models" before public release.

•The TAKE IT DOWN Act (Passed late 2025): The only major AI-specific federal legislation passed by Congress, which specifically criminalizes the generation and distribution of AI-generated non-consensual intimate imagery (deepfakes).

3. China: Proactive, Targeted Administrative Controls

China does not rely on a single, comprehensive AI statute. Instead, the Cyberspace Administration of China (CAC) regulates the industry through highly targeted administrative rules:

•Generative AI Registration: Under the 2023 Interim Measures for Generative AI Services, any developer deploying a public-facing AI system must register its algorithms and pass a safety assessment with the CAC.

•Labeling and Watermarking: Effective September 2025, the Measures for Labeling of AI-Generated Synthetic Content (backed by national standard GB 45438-2025) strictly mandates that all AI-generated content carry both visible watermarks and embedded, invisible metadata.

•Conversational & Anthropomorphic AI (July 15, 2026): China introduced the world's first rules specifically targeting AI companion chatbots. These rules mandate that "anthropomorphic" chatbots must clearly state they are artificial, restrict minors from using virtual intimate partner services, and implement "addiction-prevention" measures.

•Intelligent Agents: As of July 15, 2026, China also established a dedicated standardized framework regulating autonomous AI agents.

4. Broader Accountability: Privacy, IP, and Civil Liability

Beyond AI-specific statutes, the development of systems like me is governed by foundational legal systems:

•Intellectual Property (Copyright Law): A massive wave of high-profile lawsuits by authors, artists, and publishers is currently testing the limits of the "Fair Use" doctrine. Courts are actively determining whether tech companies have the right to scrape copyrighted web data to train LLMs without compensation.

•Data Privacy: Regulations like Europe's GDPR and California's CCPA/CPRA govern how my developers collect, store, and process training data. Users have a "right to be forgotten," which forces AI developers to find ways to extract or block personal data from being recalled by trained models.

•Algorithmic Discrimination: Under civil rights and employment laws, organizations deploying AI to make critical life decisions (such as hiring, loan approvals, or housing) are held legally liable if those algorithms perpetuate bias or discriminate against protected classes.

Citations (20)

1whitecase.comCompanies
whitecase.com
Backs: “The Global Gold Standard: The EU AI Act. The European Union’s AI Act is the most comprehensive, binding, and globally influential piece of AI legislation. It regulates AI based on its potential to cause harm, with severe penalties of up to €35 million or 7% of global annual turnover for non-compliance.”
2europa.euGovernment
europa.eu
Backs: “It regulates AI based on its potential to cause harm, with severe penalties of up to €35 million or 7% of global annual turnover for non-compliance.”
3brightdefense.comCompanies
brightdefense.com
Backs: “It regulates AI based on its potential to cause harm, with severe penalties of up to €35 million or 7% of global annual turnover for non-compliance. General-Purpose AI (GPAI) Rules: Large language models and foundation models are categorized as "GPAI models".”
4europa.euGovernment
europa.eu
Backs: “General-Purpose AI (GPAI) Rules: Large language models and foundation models are categorized as "GPAI models". Under rules that took effect in August 2025, developers must maintain rigorous technical documentation, comply with EU copyright law, publish detailed summaries of their training data, and share information…”
5artificialintelligenceact.euNonprofits and think tanks
artificialintelligenceact.eu
Backs: “General-Purpose AI (GPAI) Rules: Large language models and foundation models are categorized as "GPAI models". Under rules that took effect in August 2025, developers must maintain rigorous technical documentation, comply with EU copyright law, publish detailed summaries of their training data, and share information…”
6dataprotectionreport.comNot yet classified
dataprotectionreport.com
Backs: “Strict Transparency (Enforced August 2, 2026): The implementation of Article 50 mandates that users must be clearly informed when they are interacting with an AI (like a chatbot). Furthermore, providers must use invisible, machine-readable watermarks to label synthetic text, images, and video as AI-generated.”
7wikipedia.orgEncyclopedias and referencequality 0.83
wikipedia.org
Backs: “Instead, regulation is split between a "patchwork" of strict state-level laws and a federal push focused on global competitiveness and deregulation. State-Level Regulations (The Primary Driver of Binding Rules)”
8verifywise.aiCompanies
verifywise.ai
Backs: “State-Level Regulations (The Primary Driver of Binding Rules). Because Congress has not passed overarching AI safety legislation, states have stepped in: California’s Frontier AI Rules (In effect Jan 1, 2026): Under the Transparency in Frontier AI Act (SB 53), developers of "frontier models" (trained using more than…”
9wsgr.comCompanies
wsgr.com
Backs: “State-Level Regulations (The Primary Driver of Binding Rules). Because Congress has not passed overarching AI safety legislation, states have stepped in: California’s Frontier AI Rules (In effect Jan 1, 2026): Under the Transparency in Frontier AI Act (SB 53), developers of "frontier models" (trained using more than…”
10gc.aiCompanies
gc.ai
Backs: “State-Level Regulations (The Primary Driver of Binding Rules). Because Congress has not passed overarching AI safety legislation, states have stepped in: California’s Frontier AI Rules (In effect Jan 1, 2026): Under the Transparency in Frontier AI Act (SB 53), developers of "frontier models" (trained using more than…”
11dglaw.comNot yet classified
dglaw.com
Backs: “California's Training Data Disclosures (In effect Jan 1, 2026): AB 2013 (The Training Data Transparency Act) requires generative AI developers to publish high-level summaries on their websites detailing the data used to train, fine-tune, or test their models—specifically disclosing the use of personal data,…”
12goodwinlaw.comNot yet classified
goodwinlaw.com
Backs: “California's Training Data Disclosures (In effect Jan 1, 2026): AB 2013 (The Training Data Transparency Act) requires generative AI developers to publish high-level summaries on their websites detailing the data used to train, fine-tune, or test their models—specifically disclosing the use of personal data,…”
13drata.comCompanies
drata.com
Backs: “California's Watermarking Mandates (In effect Aug 2, 2026): Under SB 942 (the California AI Transparency Act), public generative AI services with over 1 million monthly users must offer free AI detection tools and embed latent provenance metadata inside generated media. Colorado & Texas: Texas’s TRAIGA (HB 149) went…”
14seyfarth.comNot yet classified
seyfarth.com
Backs: “Meanwhile, Colorado's heavily debated SB 24-205 was replaced in mid-2026 by SB 189, a streamlined law taking effect on January 1, 2027, that focuses on transparency and consumer disclosures regarding automated decision-making. Federal Level: Deregulation and National Preemption”
15casrai.orgNonprofits and think tanks
casrai.org
Backs: “China: Proactive, Targeted Administrative Controls. China does not rely on a single, comprehensive AI statute. Instead, the Cyberspace Administration of China (CAC) regulates the industry through highly targeted administrative rules:”
16securiti.aiNot yet classified
securiti.ai
Backs: “Generative AI Registration: Under the 2023 Interim Measures for Generative AI Services, any developer deploying a public-facing AI system must register its algorithms and pass a safety assessment with the CAC. Labeling and Watermarking: Effective September 2025, the Measures for Labeling of AI-Generated Synthetic…”
17deep-lex.comNot yet classified
deep-lex.com
Backs: “Labeling and Watermarking: Effective September 2025, the Measures for Labeling of AI-Generated Synthetic Content (backed by national standard GB 45438-2025) strictly mandates that all AI-generated content carry both visible watermarks and embedded, invisible metadata. Conversational & Anthropomorphic AI (July 15,…”
18alexmakosz.caBlogs and newsletters
alexmakosz.ca
Backs: “Labeling and Watermarking: Effective September 2025, the Measures for Labeling of AI-Generated Synthetic Content (backed by national standard GB 45438-2025) strictly mandates that all AI-generated content carry both visible watermarks and embedded, invisible metadata. Conversational & Anthropomorphic AI (July 15,…”
19rimonlaw.comNot yet classified
rimonlaw.com
Backs: “Conversational & Anthropomorphic AI (July 15, 2026): China introduced the world's first rules specifically targeting AI companion chatbots. These rules mandate that "anthropomorphic" chatbots must clearly state they are artificial, restrict minors from using virtual intimate partner services, and implement…”
20trustarc.comNot yet classified
trustarc.com
Backs: “Algorithmic Discrimination: Under civil rights and employment laws, organizations deploying AI to make critical life decisions (such as hiring, loan approvals, or housing) are held legally liable if those algorithms perpetuate bias or discriminate against protected classes. 1.”