Canonical policy question · Frontier safety

Should frontier models face mandatory evaluations?

Evaluation language is widespread in policy texts. Thresholds, evaluators and legal consequences remain contested.

Current statecontested

Policy asks

Tracked outcomes

Require public disclosure rules for foundation-model training data, testing and operations

Direct a federal regulator to establish public-information requirements covering training data, documentation, testing, inference-time collection and model operations for covered foundation models.

proposed

Require lifecycle safety governance for autonomous AI agents

Treat safety, reliability and trustworthiness as requirements across research, deployment and promotion of AI agents with autonomous capabilities.

in-force

Require large frontier developers to publish catastrophic-risk safety frameworks

Require covered large frontier developers to write, implement, comply with and publish a frontier AI framework covering catastrophic-risk assessment, mitigations and related disclosures.

adopted

Use voluntary AI risk-management frameworks for organisational governance

Adopt voluntary frameworks such as the NIST AI RMF to manage trustworthiness risks without treating the framework text as a statute.

adopted

Offer a voluntary code of practice for general-purpose AI Act compliance

Provide a multi-stakeholder code that GPAI providers may use to demonstrate compliance with AI Act transparency, copyright and systemic-risk duties.

Policy vehicles

Where the asks appear

bill · Introduced

H.R. 8094 · AI Foundation Model Transparency Act of 2026

A House bill directing the FTC to establish public disclosure requirements for covered foundation models.

guidance · Issued

Implementation Opinions on Standardized Application and Innovative Development of AI Agents

Joint policy guidance defining AI agents, setting four principles, and organising work on technology foundations, classified safety governance, nineteen application scenes and an innovation ecology.

regulation · In force

California SB 53 · Transparency in Frontier Artificial Intelligence Act

Chaptered California statute requiring large frontier developers to publish frontier AI frameworks, transparency reports and critical-incident pathways.

guidance · Published (voluntary)

NIST AI Risk Management Framework

US federal voluntary framework for managing AI trustworthiness risks across design, development, use and evaluation.

guidance · Published

EU General-Purpose AI Code of Practice

Voluntary multi-stakeholder code detailing AI Act duties for providers of general-purpose AI models.

regulation · In force

Korea AI Basic Act and Enforcement Decree

South Korea’s comprehensive AI statute combining industry promotion with transparency, safety and high-impact duties.

Policy lineage

Events and deadlines

2023-01-26
NIST releases AI Risk Management Framework 1.0
NIST published the voluntary AI Risk Management Framework.
2025-07-10
EU GPAI Code of Practice published
The Commission announced receipt of the final General-Purpose AI Code of Practice.
2025-09-29
California SB 53 chaptered into law
SB 53 was approved by the Governor and chaptered as Chapter 138, Statutes of 2025.
2026-01-01
California frontier AI transparency duties take effect
Chapter 25.1 of the Business and Professions Code became effective.
2026-01-22
Korea AI Basic Act enters into force
The AI Basic Act and Enforcement Decree took effect, with a grace period of at least one year.
2026-03-26
AI Foundation Model Transparency Act introduced in the House
H.R. 8094 entered the 119th Congress and was referred for committee consideration.
2026-05-08
China issues national policy direction for AI agents
CAC, NDRC and MIIT jointly issued implementation opinions for AI agents.

Evidence ledger

Explained source passages

Direct sourceOpen evidence record

H.R. 8094 directs the FTC to establish foundation-model disclosure rules

The bill would require regulations covering training data, documentation, testing, inference-time collection and model operations.

Establish requirements for covered entities with regard to foundation models to improve transparency of training data, documentation, testing, data collection during inference, and operations of foundation models, before commercial deployment and during the lifecycle of the system.

Congress.gov · H.R. 8094 · Section 2(a)(1)(A)

Direct sourceOpen evidence record

China’s agent guidance links development to lifecycle governance

The guidance defines agents by autonomous capabilities and places safety requirements across research, deployment and promotion.

坚持安全可控，将智能体安全、可靠、可信作为发展的底线要求，贯穿智能体技术研发、应用部署与推广的全过程，切实防范系统性风险。

Cyberspace Administration of China · Implementation Opinions · Section I guiding principles

Direct sourceOpen evidence record

China’s agent guidance defines AI agents by five capabilities

The opinions define an AI agent as an intelligent system with autonomous perception, memory, decision, interaction and execution capabilities.

智能体是具备自主感知、记忆、决策、交互与执行能力的智能系统，是人工智能产品及服务的重要形态。

Cyberspace Administration of China · Implementation Opinions · lead definition

Direct sourceOpen evidence record

China’s agent guidance keeps final decision rights with users

Item 6 requires that users retain informed consent and final decision rights over agent autonomous decisions, and that agent actions stay inside user authorisation.

确保用户对智能体自主决策享有知情权和最终决策权，智能体执行操作不得超出用户授权范围。

Cyberspace Administration of China · Implementation Opinions · Section III · item 6

Direct sourceOpen evidence record

China’s agent guidance sets classified and graded governance

Item 11 directs cyberspace and sector authorities to set open scenes in sensitive and key industries, with filing, testing and problem-product recall measures under applicable law and standards.

对于敏感领域及重点行业，由网信部门联合行业主管部门确定开放场景，根据相关法律法规、监管要求和安全防护标准，实行备案、检测、问题产品召回等管理措施。

Cyberspace Administration of China · Implementation Opinions · Section III · item 11

Direct sourceOpen evidence record

California requires large frontier developers to publish a frontier AI framework

Business and Professions Code section 22757.12(a) requires covered developers to write, implement, comply with and publish a frontier AI framework.

A large frontier developer shall write, implement, comply with, and clearly and conspicuously publish on its internet website a frontier AI framework that applies to the large frontier developer’s frontier models and describes how the large frontier developer approaches all of the following:

California Legislative Information · Cal. Bus. & Prof. Code § 22757.12(a)

Direct sourceOpen evidence record

California’s frontier AI transparency chapter took effect on 1 January 2026

The chaptered statute records an effective date of 1 January 2026.

(Added by Stats. 2025, Ch. 138, Sec. 2. (SB 53) Effective January 1, 2026.)

California Legislative Information · Cal. Bus. & Prof. Code § 22757.10 historical note

Direct sourceOpen evidence record

NIST states the AI RMF is intended for voluntary use

NIST’s AI RMF overview describes the framework as voluntary and designed to improve trustworthiness risk management.

The NIST AI Risk Management Framework (AI RMF) is intended for voluntary use and to improve the ability to incorporate trustworthiness considerations into the design, development, use, and evaluation of AI products, services, and systems.

National Institute of Standards and Technology · NIST AI RMF overview page · Overview of the AI RMF

Direct sourceOpen evidence record

Commission received the final GPAI Code of Practice on 10 July 2025

The Commission announcement records receipt of the final voluntary Code covering transparency, copyright and safety/security chapters.

The European Commission has received the final version of the General-Purpose AI Code of Practice, a voluntary tool developed by 13 independent experts, with input from over 1,000 stakeholders, including model providers, small and medium-sized enterprises, academics, AI safety experts, rightsholders, and civil society organisations.

European Commission · Commission announcement · 10 July 2025

Direct sourceOpen evidence record

Korea’s AI Basic Act took effect on 22 January 2026

MSIT states the AI Basic Act and Enforcement Decree come into force on 22 January 2026.

The Ministry of Science and ICT (MSIT, Deputy Prime Minister and Minister: Bae Kyung-hoon) announced that the Basic Act on the Development of Artificial Intelligence and the Establishment of a Trustworthy Foundation (“AI Basic Act”) comes into effect on January 22.

Ministry of Science and ICT (Korea) · MSIT press release · lead

Intelligence records

Reported developments

China · Governance signal

China sets a policy direction for autonomous AI agents

New guidance frames agents as systems with perception, memory, decision and execution capabilities, pairing deployment goals with standards, certification and lifecycle governance.

United States · Bill introduced

US proposal would put foundation-model disclosure under the FTC

H.R. 8094 asks the Federal Trade Commission to establish public-information requirements covering foundation-model training data and algorithms. The vehicle remains at the introduced stage.

California · State statute in force

California requires large frontier developers to publish safety frameworks

SB 53, chaptered in September 2025, took effect on 1 January 2026 as Business and Professions Code Chapter 25.1. Covered large frontier developers must publish a frontier AI framework and related transparency reports.

EU · Compliance tool

EU publishes a voluntary GPAI Code of Practice for AI Act duties

On 10 July 2025 the Commission announced receipt of the final General-Purpose AI Code of Practice. The code covers transparency and copyright for all GPAI providers and safety/security for systemic-risk models.

United States · Standards baseline

NIST keeps the AI Risk Management Framework voluntary

NIST’s AI RMF, released on 26 January 2023, remains a voluntary organisational tool. Later profiles address generative AI and critical infrastructure without converting the core framework into a statute.