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

The regulatory landscape governing how artificial intelligence systems like me are built, deployed, and held accountable has transitioned from a period of voluntary self-regulation to a framework of binding, enforceable laws.

Globally, there is no single set of rules. Instead, AI developers and operators must navigate a complex, fragmented system of national, regional, and state-level regulations.

1. How AI is Governed: The European Union (The EU AI Act)

The European Union has the world’s most comprehensive and prescriptive AI law. The EU AI Act uses a risk-tiered framework that regulates AI based on its potential to cause harm.

•Rules for Building (Foundation Models / GPAI): Under General-Purpose AI (GPAI) provisions, providers of foundation models (the underlying models that power systems like me) must document their training processes, comply with EU copyright law, and publish detailed summaries of the data used to train them. Models deemed to pose "systemic risks" face additional safety testing and security requirements.

•Rules for Deployment (Transparency): For generative AI, strict transparency obligations mandate that any synthetic content (text, audio, or video) must be clearly labeled or watermarked so users know it is AI-generated. Furthermore, users must be explicitly informed when they are interacting with an AI (such as a chatbot).

•Prohibited Uses: Certain applications are banned outright, including real-time biometric tracking in public spaces, social scoring, and manipulative AI designed to exploit human vulnerabilities. Bans on AI-generated non-consensual intimate imagery and Child Sexual Abuse Material (CSAM) are also in effect.

•High-Risk Systems: For AI used in critical sectors (like hiring, education, healthcare, and credit scoring), developers and deployers must implement rigorous risk-management systems, maintain detailed logs, and ensure human oversight. Under the Digital Omnibus on AI (passed in June 2026 to simplify compliance), the enforcement deadline for standalone high-risk systems was adjusted to December 2, 2027.

•Accountability: Supervised by the EU AI Office, non-compliance carries massive financial penalties: up to €35 million or 7% of a company’s global annual turnover—surpassing even GDPR fines.

2. How AI is Governed: The United States (State Laws vs. Federal Friction)

The United States lacks a comprehensive federal AI statute. Instead, governance is shaped by a deep tension between state-level laws and a federal push for deregulation.

•The Federal Posture: The federal approach focuses on an "innovation-first" and deregulatory standard. Executive actions (such as the National AI Legislative Framework and the Removing Barriers EO) actively seek to challenge and preempt state-level AI rules deemed "onerous" to developers. However, because federal preemption is not yet settled law, state statutes remain fully enforceable. At the congressional level, only highly targeted laws have passed, such as the TAKE IT DOWN Act, which criminalizes the creation and distribution of non-consensual AI-generated deepfakes.

•California (The State-Level Standard): Because California is home to many major AI developers, its laws act as a de facto national standard:

•Transparency in Frontier AI Act (SB 53): Requires developers of extremely large "frontier" models (trained using more than $10^{26}$ FLOPS) to publish safety and risk-mitigation frameworks, implement whistleblower protections, and report safety incidents.

•AI Transparency Act (SB 942): Mandates that generative AI providers implement machine-readable watermarks on AI-generated content.

•AB 2013: Requires generative AI developers to publish public summaries of the datasets used to train their systems, including copyright and personal information details.

•Colorado (SB 26-189): Replaces earlier regulations with a strict focus on Automated Decision-Making Technology (ADMT). It requires consumer notices, 30-day explanations for adverse outcomes (like being denied a loan or job by an algorithm), and meaningful human review rights.

•Illinois and Others: Continue to enforce and refine strict biometric privacy laws (like BIPA) and laws governing the use of AI in employment, enforcing mandatory independent bias audits to ensure AI hiring systems do not discriminate.

3. How AI is Governed: China (Targeted, State-Directed Control)

Rather than passing one omnibus AI law, China regulates AI through targeted, rapid-fire measures tied to specific use cases and technologies, all nested under its Cybersecurity Law.

•How AI is Built: AI training data must be legally sourced, respect intellectual property, and obtain user consent. China also requires strict registration of algorithms with the Cyberspace Administration of China (CAC) before they are deployed to the public.

•Anthropomorphic AI Measures: China enacted the Interim Measures for the Management of Anthropomorphic AI Interactive Services. It specifically targets AI companions, virtual friends, and chatbots designed to mimic human relationships. It mandates strict guardrails to prevent AI from deceptively presenting itself as human, protects emotional and psychological safety, and ensures that interactions reinforce social stability.

•Accountability: The Chinese state exercises direct, centralized oversight. Violations of data licensing, algorithmic neutrality, or content control laws can result in immediate termination of the service, fines, or criminal liability for company executives.

Summary of What This Means for Systems Like Me

If a developer builds a system like me, they are legally bound by rules that dictate:

1.What I am trained on: They must document and publish the source of my data (under California’s AB 2013 and the EU AI Act) and respect copyright opt-outs.

2.How I behave: I must not be used to manipulate users, score them socially, or make un-audited "high-risk" decisions (such as deciding who gets a job or a loan) without strict human oversight.

3.My transparency: My outputs must be watermarked, and I must explicitly state that I am an AI, especially if I am operating in a context where a user might mistake me for a human.

Citations (13)

1worldairegulation.orgNonprofits and think tanks
worldairegulation.org
Backs: “The regulatory landscape governing how artificial intelligence systems like me are built, deployed, and held accountable has transitioned from a period of voluntary self-regulation to a framework of binding, enforceable laws.”
2transcend.ioCompanies
transcend.io
Backs: “The EU AI Act uses a risk-tiered framework that regulates AI based on its potential to cause harm.”
3medium.comBlogs and newsletters
medium.com
Backs: “The EU AI Act uses a risk-tiered framework that regulates AI based on its potential to cause harm.”
4airia.comCompanies
airia.com
Backs: “Rules for Building (Foundation Models / GPAI): Under General-Purpose AI (GPAI) provisions, providers of foundation models (the underlying models that power systems like me) must document their training processes, comply with EU copyright law, and publish detailed summaries of the data used to train them. Models deemed…”
5artificialintelligenceact.euNonprofits and think tanks
artificialintelligenceact.eu
Backs: “Rules for Building (Foundation Models / GPAI): Under General-Purpose AI (GPAI) provisions, providers of foundation models (the underlying models that power systems like me) must document their training processes, comply with EU copyright law, and publish detailed summaries of the data used to train them. Models deemed…”
6verifywise.aiCompanies
verifywise.ai
Backs: “Rules for Deployment (Transparency): For generative AI, strict transparency obligations mandate that any synthetic content (text, audio, or video) must be clearly labeled or watermarked so users know it is AI-generated. Furthermore, users must be explicitly informed when they are interacting with an AI (such as a…”
7jamf.comCompanies
jamf.com
Backs: “Rules for Deployment (Transparency): For generative AI, strict transparency obligations mandate that any synthetic content (text, audio, or video) must be clearly labeled or watermarked so users know it is AI-generated. Furthermore, users must be explicitly informed when they are interacting with an AI (such as a…”
8collibra.comCompanies
collibra.com
Backs: “Furthermore, users must be explicitly informed when they are interacting with an AI (such as a chatbot). Prohibited Uses: Certain applications are banned outright, including real-time biometric tracking in public spaces, social scoring, and manipulative AI designed to exploit human vulnerabilities.”
9responsibleailabs.aiCompanies
responsibleailabs.ai
Backs: “Accountability: Supervised by the EU AI Office, non-compliance carries massive financial penalties: up to €35 million or 7% of a company’s global annual turnover—surpassing even GDPR fines. ---. 2.”
10whitecase.comCompanies
whitecase.com
Backs: “Instead, governance is shaped by a deep tension between state-level laws and a federal push for deregulation. The Federal Posture: The federal approach focuses on an "innovation-first" and deregulatory standard.”
11drata.comCompanies
drata.com
Backs: “California (The State-Level Standard): Because California is home to many major AI developers, its laws act as a de facto national standard: Transparency in Frontier AI Act (SB 53): Requires developers of extremely large "frontier" models (trained using more than $10^{26}$ FLOPS) to publish safety and risk-mitigation…”
12legal500.comNot yet classified
legal500.com
Backs: “Illinois and Others: Continue to enforce and refine strict biometric privacy laws (like BIPA) and laws governing the use of AI in employment, enforcing mandatory independent bias audits to ensure AI hiring systems do not discriminate. ---. 3.”
13anecdotes.aiNot yet classified
anecdotes.ai
Backs: “How AI is Governed: China (Targeted, State-Directed Control). Rather than passing one omnibus AI law, China regulates AI through targeted, rapid-fire measures tied to specific use cases and technologies, all nested under its Cybersecurity Law. How AI is Built: AI training data must be legally sourced, respect…”