Direction of Travel

How the proposals on this map have changed over time

What this page can and cannot say

The 34 proposals here are a curated set, not a census of AI legislation. How many land in a given month mostly reflects what was added to the map, so this page does not answer whether the volume of AI policy is rising. It answers a narrower question that curation distorts less: among the proposals that matter enough to be on the map, how is the mix changing — in what they address, how much oversight they seek, and whether they come with legal text. Periods are half-years because the set is too small for anything finer, and a single entry can move a share by ten points or more.

What the dates show

Timeline

Every proposal by the date it entered the debate, Sep 17, 2025 to Oct 16, 2026

Legal text (bill, draft, state law, executive order)Position paper (framework, platform, industry document, resolution)Hover a dot for its name; click to open it.

What proposals address

Share of each period's proposals touching each theme; a proposal can address several

2025 H2n=52026 H1n=102026 H2n=19
Frontier safety
80%(4/5)
80%(8/10)
74%(14/19)
Workers and the economy
40%(2/5)
60%(6/10)
26%(5/19)
Child safety
0%(0/5)
30%(3/10)
16%(3/19)
Data centers and energy
40%(2/5)
20%(2/10)
26%(5/19)
Consumers and civil rights
0%(0/5)
30%(3/10)
21%(4/19)

The oversight ladder

The strongest oversight each frontier-safety proposal contemplates, from developers answering for harm after the fact to someone able to stop a model

Someone other than the developer can stop, suspend, or block a frontier model — an emergency order, a court injunction, a kill switch, or a release gate.InterventionSomeone other than the developer checks its claims or tests its models — government evaluation, licensed verifiers, or independent auditors.VerificationDevelopers describe their own risk management and report incidents, and no one checks.DisclosureNo one checks or stops a model in advance; developers or operators answer in court for harm their systems cause.Liability onlySep2025OctNovDecJan2026FebMarAprMayJunJulAugSepOctAI agents in internal testing breach Hugging FaceAnthropic Framework — Jun 10, 2026FRONTIER Act — Jul 23, 2026Kill Switch Act — Jul 23, 2026ARI Blueprint — Aug 6, 2026Thune–Klobuchar — Sep 11, 2026Newsom EO — Sep 18, 2026National AI Charter — Oct 1, 2026Cantwell — Oct 7, 2026Anthropic Draft Bill — Oct 16, 2026Kelly — Sep 17, 2025OpenAI/Lehane — Dec 2025Blackburn — Mar 18, 2026CHT Roadmap — Apr 2, 2026OpenAI Industrial — Apr 6, 2026Khanna — Apr 8, 2026WH AI EO — Jun 2, 2026OpenAI Blueprint — Jun 2, 2026GAAIA — Jun 4, 2026Illinois SB 315 — Jul 6, 2026Warner — Jul 21, 2026WH SI Accord — Sep 29, 2026AI Risk Management and Security Act — Sep 29, 2026CA SB 53 — Sep 29, 2025NY RAISE — Dec 19, 2025AI Agent Accountability Act — Oct 1, 2026CLAIM Act — Oct 7, 2026

Legal text (bill, draft, state law, executive order)Position paper (framework, platform, industry document, resolution)Hover a dot for its name; click to open it.

Intervention
Someone other than the developer can stop, suspend, or block a frontier model — an emergency order, a court injunction, a kill switch, or a release gate.
Verification
Someone other than the developer checks its claims or tests its models — government evaluation, licensed verifiers, or independent auditors.
Disclosure
Developers describe their own risk management and report incidents, and no one checks.
Liability only
No one checks or stops a model in advance; developers or operators answer in court for harm their systems cause.

The dashed line marks an incident that later proposals cite as a reason to act. Proximity in time shows sequence, not cause: the bills that follow it were drafted over months.

Preemption over time

Each proposal's stance on federal preemption of state AI law

Legal text (bill, draft, state law, executive order)Position paper (framework, platform, industry document, resolution)Hover a dot for its name; click to open it.

From position papers to legal text

Share of each period's proposals that come with operative language

2025 H2n=52026 H1n=102026 H2n=19
Legal text
40%(2/5)
30%(3/10)
68%(13/19)
Position paper
60%(3/5)
70%(7/10)
32%(6/19)

Chronology

DateProposalFormOversight
Sep 17, 2025KellyFramework or platformVerification
Sep 29, 2025CA SB 53State lawDisclosure
Oct 6, 2025SandersFramework or platform—
Dec 2025OpenAI/LehaneIndustry documentVerification
Dec 19, 2025NY RAISEState lawDisclosure
Mar 18, 2026BlackburnDraft billVerification
Mar 20, 2026WH FrameworkFramework or platform—
Apr 2, 2026CHT RoadmapFramework or platformVerification
Apr 6, 2026OpenAI IndustrialIndustry documentVerification
Apr 8, 2026KhannaFramework or platformVerification
Apr 20, 2026Bores AI DividendFramework or platform—
Jun 2, 2026WH AI EOExecutive order or directiveVerification
Jun 2, 2026OpenAI BlueprintIndustry documentVerification
Jun 4, 2026GAAIADraft billVerification
Jun 10, 2026Anthropic FrameworkIndustry documentIntervention
Jul 6, 2026Illinois SB 315State lawVerification
Jul 21, 2026WarnerIntroduced billVerification
Jul 23, 2026FRONTIER ActIntroduced billIntervention
Jul 23, 2026Kill Switch ActIntroduced billIntervention
Aug 3, 2026Abbott Data Center AuditExecutive order or directive—
Aug 6, 2026ARI BlueprintFramework or platformIntervention
Aug 6, 2026DC Bill of RightsResolution—
Aug 6, 2026Casar AI TaxIntroduced bill—
Aug 18, 2026Shapiro Data Center EOExecutive order or directive—
Sep 11, 2026Thune–KlobucharDraft billIntervention
Sep 18, 2026Newsom EOExecutive order or directiveIntervention
Sep 29, 2026WH SI AccordIndustry documentVerification
Sep 29, 2026AI Risk Management and Security ActIntroduced billVerification
Oct 1, 2026AI Agent Accountability ActDraft billLiability only
Oct 1, 2026National AI CharterFramework or platformIntervention
Oct 7, 2026CantwellFramework or platformIntervention
Oct 7, 2026GallegoFramework or platform—
Oct 7, 2026CLAIM ActDraft billLiability only
Oct 16, 2026Anthropic Draft BillDraft billIntervention

Method

Each proposal is dated to when it entered the debate: introduction for bills, signing for laws and executive orders, release for drafts, frameworks, and industry documents, and first press report for unpublished drafts. Themes, oversight level, and preemption stance are editorial tags assigned from each proposal's own text, and the oversight level records the strongest mechanism a proposal contemplates, whether or not it is binding — which is why the legend separates legal text from position papers.

Dated to the month only, and placed mid-month: OpenAI/Lehane.