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

The rules governing how artificial intelligence systems like me are built, deployed, and held accountable have shifted from voluntary ethical guidelines to a complex, binding web of international regulations.

Because I am a General-Purpose AI (GPAI)—a model capable of performing a wide variety of tasks—I fall under a distinct set of rules compared to narrow, single-task AI. Currently, the regulatory landscape is dominated by a major framework in Europe, a rapidly growing patchwork of state-level laws in the United States, and strict centralized controls in Asia.

1. Building AI: Rules on Development and Training Data

Before a model is ever deployed, developers face strict rules regarding how models are trained and what data they can ingest.

•Training Data Transparency (U.S. & EU):

•Under California’s AB 2013 (effective January 1, 2026), developers of generative AI must publish detailed public summaries of their training datasets, outlining data sources, intellectual property info, and personal data details.

•The EU AI Act requires GPAI developers to provide detailed technical documentation and a publicly available summary of the content used to train their models.

•Copyright and Fair Use: While not "AI-specific" legislation, ongoing landmark copyright lawsuits brought by authors, artists, and media companies are actively shaping the legal boundaries of what data AI developers can legally use for training without a license.

•Frontier Model Safety Measures: For "frontier" models—typically defined by huge compute thresholds (such as training with more than $10^{26}$ FLOPS)—the regulatory burden is much higher:

•California's SB 53 (Transparency in Frontier AI Act), which took effect in January 2026, forces developers of massive models to publish risk frameworks, report critical safety incidents, and implement whistleblower protections.

•Illinois enacted the Artificial Intelligence Safety Measures Act in July 2026, imposing catastrophic-risk management, safety testing, and incident-reporting duties on frontier model developers.

2. Deploying AI: Rules on Use, Interaction, and Consumer Protection

When AI is integrated into software and interacts with the public, deployment rules focus heavily on safety, transparency, and discrimination.

•Prohibited AI Practices (European Union): Under the EU AI Act, certain uses of AI are entirely banned. These include "unacceptable-risk" applications like real-time biometric surveillance in public spaces, social scoring, or systems designed to manipulate human behavior to cause harm.

•Interacting with Humans (Watermarking and Disclosure): Many jurisdictions now legally require systems like me to make it obvious to users that they are talking to an AI, not a human:

•California’s SB 942 (AI Transparency Act) requires AI providers to disclose when content (text, audio, images, or video) is AI-generated, often via digital watermarking.

•Under the EU AI Act (Article 50), generative AI providers must ensure synthetic content is marked as machine-generated.

•Automated Decision-Making Technology (ADMT): When AI is used to make "consequential decisions"—such as who gets a job, a loan, housing, or medical treatment—regulations are especially tight:

•Colorado's SB 189 (which repealed and replaced the older Colorado AI Act in May 2026) requires organizations using AI for automated decisions to provide consumers with clear pre-use disclosures, explanations of adverse outcomes, and rights to human review.

•Similar targeted laws exist in Connecticut, Texas, and several major municipalities (like New York City's automated employment decision law).

3. Holding AI Accountable: Enforcement, Liability, and Fines

Who is responsible when an AI system hallucinates incorrect medical advice, discriminates against job applicants, or commits cybercrime?

•The EU AI Act & The European AI Office: The EU's central AI Office has full enforcement powers over GPAI models. They can request internal technical documentation, evaluate models, force corrective measures, and issue historic fines of up to 7% of a company’s global annual turnover (or €35 million, whichever is higher) for major compliance failures.

•The U.S. Federal/State Tension:

•At the U.S. federal level, there is still no comprehensive national AI law. However, the federal government has targeted specific harms, such as the TAKE IT DOWN Act, which criminalizes the creation and distribution of AI-generated non-consensual intimate imagery (deepfakes).

•The executive branch has largely shifted toward a "pro-innovation" stance. Under President Trump's Executive Order 14409 (June 2026), the administration established a voluntary secure-deployment framework for NSA-designated "covered frontier models" and directed agencies to prioritize AI for cyber-defense. The White House has also pushed for federal preemption of state-level laws, setting up ongoing legal battles over who has the final say on holding AI companies liable.

•China’s Generative AI Regulations: Managed strictly by the Cyberspace Administration of China (CAC), developers of generative AI must register their algorithms through a formal filing system, conduct security assessments, and ensure that AI outputs align with socialist values and do not subvert state power.

Summary

Ultimately, if an organization wants to build or deploy a model like me today, they must navigate a dual reality: in Europe, they must comply with the central, strict, risk-based mandates of the EU AI Act; in the United States, they must navigate a complex, state-driven legal minefield while federal agencies focus primarily on cybersecurity and national competitiveness.

Citations (12)

1verifywise.aiCompanies
verifywise.ai
Backs: “Training Data Transparency (U.S. & EU): Under California’s AB 2013 (effective January 1, 2026), developers of generative AI must publish detailed public summaries of their training datasets, outlining data sources, intellectual property info, and personal data details. The EU AI Act requires GPAI developers to provide…”
2alicelabs.aiCompanies
alicelabs.ai
Backs: “The EU AI Act requires GPAI developers to provide detailed technical documentation and a publicly available summary of the content used to train their models. Copyright and Fair Use: While not "AI-specific" legislation, ongoing landmark copyright lawsuits brought by authors, artists, and media companies are actively…”
3europa.euGovernment
europa.eu
Backs: “The EU AI Act requires GPAI developers to provide detailed technical documentation and a publicly available summary of the content used to train their models. Copyright and Fair Use: While not "AI-specific" legislation, ongoing landmark copyright lawsuits brought by authors, artists, and media companies are actively…”
4hinshawlaw.comCompanies
hinshawlaw.com
Backs: “Illinois enacted the Artificial Intelligence Safety Measures Act in July 2026, imposing catastrophic-risk management, safety testing, and incident-reporting duties on frontier model developers. ---. 2.”
5responsibleailabs.aiCompanies
responsibleailabs.ai
Backs: “Under the EU AI Act (Article 50), generative AI providers must ensure synthetic content is marked as machine-generated. Automated Decision-Making Technology (ADMT): When AI is used to make "consequential decisions"—such as who gets a job, a loan, housing, or medical treatment—regulations are especially tight:”
6artificialintelligenceact.euNonprofits and think tanks
artificialintelligenceact.eu
Backs: “Under the EU AI Act (Article 50), generative AI providers must ensure synthetic content is marked as machine-generated. Automated Decision-Making Technology (ADMT): When AI is used to make "consequential decisions"—such as who gets a job, a loan, housing, or medical treatment—regulations are especially tight:”
7seyfarth.comNot yet classified
seyfarth.com
Backs: “Automated Decision-Making Technology (ADMT): When AI is used to make "consequential decisions"—such as who gets a job, a loan, housing, or medical treatment—regulations are especially tight: Colorado's SB 189 (which repealed and replaced the older Colorado AI Act in May 2026) requires organizations using AI for…”
8troutmanprivacy.comNot yet classified
troutmanprivacy.com
Backs: “Automated Decision-Making Technology (ADMT): When AI is used to make "consequential decisions"—such as who gets a job, a loan, housing, or medical treatment—regulations are especially tight: Colorado's SB 189 (which repealed and replaced the older Colorado AI Act in May 2026) requires organizations using AI for…”
9consumerfinancemonitor.comNot yet classified
consumerfinancemonitor.com
Backs: “Automated Decision-Making Technology (ADMT): When AI is used to make "consequential decisions"—such as who gets a job, a loan, housing, or medical treatment—regulations are especially tight: Colorado's SB 189 (which repealed and replaced the older Colorado AI Act in May 2026) requires organizations using AI for…”
10whitecase.comCompanies
whitecase.com
Backs: “Similar targeted laws exist in Connecticut, Texas, and several major municipalities (like New York City's automated employment decision law). ---. 3. Holding AI Accountable: Enforcement, Liability, and Fines”
11drata.comCompanies
drata.com
Backs: “The executive branch has largely shifted toward a "pro-innovation" stance. Under President Trump's Executive Order 14409 (June 2026), the administration established a voluntary secure-deployment framework for NSA-designated "covered frontier models" and directed agencies to prioritize AI for cyber-defense.”
12wikipedia.orgEncyclopedias and referencequality 0.83
wikipedia.org
Backs: “Under President Trump's Executive Order 14409 (June 2026), the administration established a voluntary secure-deployment framework for NSA-designated "covered frontier models" and directed agencies to prioritize AI for cyber-defense. The White House has also pushed for federal preemption of state-level laws, setting up…”