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

The rules governing how artificial intelligence systems—specifically large language models (LLMs) and generative AI models—are built, deployed, and held accountable are undergoing a rapid transition. What was once a landscape dominated by voluntary guidelines has shifted into an era of mandatory, binding regulations, active enforcement, and a patchwork of national and state-level laws.

These rules generally fall into three key buckets: global/multilateral regulations, U.S. state and federal policies, and industry-specific requirements.

1. The European Union: The EU AI Act (The Global Benchmark)

The EU AI Act is the world’s most comprehensive binding AI law, and its enforcement has reached critical milestones.

•General-Purpose AI (GPAI) Obligations: For foundational models like the ones powering me, strict rules have been in effect. GPAI developers must:

•Maintain extensive technical documentation.

•Provide transparency to downstream developers who build applications on top of the models.

•Establish clear policies to comply with EU copyright law.

•Publish a detailed summary of the content used to train the model.

•Full Enforcement and Fines: The European Commission’s AI Office holds full enforcement and fining powers. Non-compliance can result in massive financial penalties—up to €15 million or 3% of a company's global turnover.

•Transparency Requirements (Article 50): Interactive AI systems (like chatbots) must explicitly disclose to users that they are interacting with a machine. Additionally, developers of generative AI must ensure that synthetic text, audio, and video are technically marked and identifiable (e.g., through watermarking or metadata provenance).

•Banned Practices: High-risk practices—such as untargeted scraping of facial images from the internet, emotion recognition in workplaces, and certain types of social scoring—are strictly prohibited.

2. United States: Federal Policy and Evolving Executive Action

At the federal level, the United States lacks a singular, comprehensive AI law passed by Congress. Instead, regulation is driven by executive actions, existing agency powers, and voluntary frameworks.

•Federal Executive Orders: A December 2025 Executive Order ("Ensuring a National Policy Framework for Artificial Intelligence") directed federal agencies to maintain U.S. AI dominance while establishing a "minimally burdensome" national standards framework. Crucially, the federal government is actively seeking to coordinate and, where possible, preempt conflicting or overly restrictive state-level AI regulations through litigation and funding conditions.

•Federal Agency Enforcement: Existing regulators use their established statutory authority to govern AI:

•The Federal Trade Commission (FTC) actively polices the market under its authority to ban "unfair or deceptive practices." This includes holding companies accountable for deceptive advertising of AI capabilities, algorithmic bias, and the unauthorized scraping of consumer data.

•The Department of Justice (DOJ) and Equal Employment Opportunity Commission (EEOC) monitor and prosecute civil rights violations when automated algorithms cause discriminatory outcomes in hiring, housing, or lending.

•NIST Risk Management: The National Institute of Standards and Technology (NIST) AI Risk Management Framework (RMF) serves as the primary federal benchmark for how organizations should measure, manage, and mitigate AI risks during design and deployment.

3. United States: A Complex State-Level Patchwork

In the absence of a federal law, individual states have stepped in with highly active legislative frameworks.

•California: The state has implemented some of the nation’s strictest rules.

•AB 2013 (AI Training Data Transparency Act): Mandates that generative AI developers publish comprehensive summaries of their training datasets, listing data sources, types, and whether any personal or intellectual property data was included.

•SB 942 (AI Transparency Act): Requires AI providers to implement system-level watermarking and digital disclosures so that users can verify when content is AI-generated.

•SB 53 (Transparency in Frontier AI Act): Subjects developers of "frontier models" (extremely large systems trained above a threshold of $10^{26}$ FLOPS) to mandatory risk reporting, whistleblower protections, and severe penalties for failing to report safety incidents.

•Colorado: Having initially passed a sweeping AI Act, Colorado updated its approach with SB 26-189, which shifts focus toward consumer-facing protections. It mandates pre-use consumer notices, 30-day explanations for adverse algorithmic decisions, and meaningful human review of automated outcomes.

•Employment and Bias Laws: States like Illinois, New York, and Texas have enacted sector-specific rules targeted at AI used in the workforce. These laws require regular bias audits, employee disclosures, and the ability for job applicants to opt-out of fully automated employment decisions.

4. International Regulations (Beyond the US and EU)

•China: China enforces specific, rigorous rules targeting Generative AI. Under its Interim Measures for the Management of Generative AI Services, developers must register their algorithms with the Cyberspace Administration of China (CAC), ensure training data is legally obtained, label synthetic content, and adhere to strict content moderation standards to prevent the generation of restricted materials.

•Other Nations: Countries such as South Korea, Vietnam, Kazakhstan, and Canada have begun putting dedicated national AI laws or regulatory sandboxes into practice, centering on transparency, data privacy, and mandatory risk assessments.

Summary: How Accountability is Enforced

If a system like me causes harm, violates privacy, or acts deceptively, accountability is established through several key mechanisms:

1.Developer Liability: AI creators can be sued under existing product liability, consumer protection, or civil rights laws if their systems exhibit severe bias or cause financial or physical harm.

2.Mandatory Audits: In sectors like employment and finance, developers and deployers are legally required to undergo regular third-party audits to prove their models are not producing discriminatory outcomes.

3.Data Provenance: Under laws like the EU AI Act and California’s AB 2013, the developers behind major models must legally "show their work" regarding training data, making them vulnerable to copyright litigation if they cannot prove compliance.

Citations (12)

1transcend.ioCompanies
transcend.io
Backs: “The rules governing how artificial intelligence systems—specifically large language models (LLMs) and generative AI models—are built, deployed, and held accountable are undergoing a rapid transition. What was once a landscape dominated by voluntary guidelines has shifted into an era of mandatory, binding regulations,…”
2collibra.comCompanies
collibra.com
Backs: “The rules governing how artificial intelligence systems—specifically large language models (LLMs) and generative AI models—are built, deployed, and held accountable are undergoing a rapid transition. What was once a landscape dominated by voluntary guidelines has shifted into an era of mandatory, binding regulations,…”
3alicelabs.aiCompanies
alicelabs.ai
Backs: “General-Purpose AI (GPAI) Obligations: For foundational models like the ones powering me, strict rules have been in effect. GPAI developers must:”
4brightdefense.comCompanies
brightdefense.com
Backs: “General-Purpose AI (GPAI) Obligations: For foundational models like the ones powering me, strict rules have been in effect. GPAI developers must:”
5europa.euGovernment
europa.eu
Backs: “Full Enforcement and Fines: The European Commission’s AI Office holds full enforcement and fining powers. Non-compliance can result in massive financial penalties—up to €15 million or 3% of a company's global turnover.”
6europa.euGovernment
europa.eu
Backs: “Banned Practices: High-risk practices—such as untargeted scraping of facial images from the internet, emotion recognition in workplaces, and certain types of social scoring—are strictly prohibited. ---. 2. United States: Federal Policy and Evolving Executive Action”
7verifywise.aiCompanies
verifywise.ai
Backs: “United States: Federal Policy and Evolving Executive Action. At the federal level, the United States lacks a singular, comprehensive AI law passed by Congress. Instead, regulation is driven by executive actions, existing agency powers, and voluntary frameworks.”
8drata.comCompanies
drata.com
Backs: “United States: Federal Policy and Evolving Executive Action. At the federal level, the United States lacks a singular, comprehensive AI law passed by Congress. Instead, regulation is driven by executive actions, existing agency powers, and voluntary frameworks.”
9gunder.comLow provenance
gunder.com
Backs: “Federal Executive Orders: A December 2025 Executive Order ("Ensuring a National Policy Framework for Artificial Intelligence") directed federal agencies to maintain U.S. AI dominance while establishing a "minimally burdensome" national standards framework. Crucially, the federal government is actively seeking to…”
10legal500.comNot yet classified
legal500.com
Backs: “The Department of Justice (DOJ) and Equal Employment Opportunity Commission (EEOC) monitor and prosecute civil rights violations when automated algorithms cause discriminatory outcomes in hiring, housing, or lending. NIST Risk Management: The National Institute of Standards and Technology (NIST) AI Risk Management…”
11holisticai.comCompanies
holisticai.com
Backs: “These laws require regular bias audits, employee disclosures, and the ability for job applicants to opt-out of fully automated employment decisions. ---. 4. International Regulations (Beyond the US and EU)”
12verifywise.aiCompanies
verifywise.ai
Backs: “Other Nations: Countries such as South Korea, Vietnam, Kazakhstan, and Canada have begun putting dedicated national AI laws or regulatory sandboxes into practice, centering on transparency, data privacy, and mandatory risk assessments. ---. Summary: How Accountability is Enforced”