Free instruments

The public record of AI governance, free and open

41 instruments on 7 shelves, no sign-up and no email wall. The law as it stands, what regulators have actually done, who is hiring, what each AI product does with your data, what an agent may do and where it went wrong, who governs the robot, who pays for the grid, and what frontier models have actually done outside their boundaries, each claim carrying the source it came from and the date it was checked. Study aids, not legal advice: the official texts are always authoritative.

Your question. A place to start.

What are you working on?

Search the tools by task, regulation or question. Open an instrument and work with its evidence.

Find your instrument

Focus your search

41 of 41 instruments

Law and enforcement

What the law says, what regulators have actually done, and whether it applies to you.

EU AI Act Explorer

What does the EU AI Act actually say, article by article?

The whole of Regulation (EU) 2024/1689 as one connected map, consolidated with the Digital Omnibus (Regulation (EU) 2026/1744) applied.

What you can do here

Every article, recital, annex and definition has its own page, carrying the full official text.
Each page lists what the provision cites, what cites it, and the recitals and annexes behind it.
Amended provisions carry a marker and list what the Digital Omnibus changed.
6 guided reading paths through the Regulation, by role and by question.

Open the Explorer

The EU AI Act Enforcement Tracker

What has a European regulator actually done to an AI system?

The public record of AI enforcement in Europe. Every fine, corrective order, ban, warning and open investigation against an AI system, dated, sourced, and linked to the provision it rests on.

What you can do here

One indexable page per enforcement action, under a permanent record identifier that is never reused, so a citation still resolves in five years.
Appeal status is a first class field: an annulled fine is struck through and excluded from every total, which is how the tracker avoids reporting the 15 million euro Italian fine against OpenAI that a court set aside in March 2026.
A page for each of the 9 enforceable provisions of the AI Act, carrying its penalty tier, the date it binds, and the actions citing it, linked both ways with the Explorer.
Who enforces it, which member states have designated an authority, and the whole dataset free as JSON, CSV and an Atom feed under CC BY 4.0.

Open the tracker

EU AI Act Scope Wizard

Is my AI system in scope of the EU AI Act, and at which risk tier?

The question every team asks first, answered against the consolidated text: is this system in scope, and at which risk tier?

What you can do here

10 questions across definition, Article 2 exclusions, territory, general-purpose models, Article 5 prohibitions, Annex I, Annex III, the Article 6(3) filter and Article 50 transparency.
Ten possible outcomes, each carrying the provisions it rests on quoted in full, the date it starts to bind, and the point where the answer stops being safe to rely on.
Follows the Digital Omnibus: the two new Article 5 prohibitions, the narrowed safety-component test, and the December 2027 and August 2028 high-risk dates.
It asks which section of Annex I applies, because Article 2(2) leaves only Article 6(1), Article 60a and Articles 102 to 112 applicable to a Section B product.
Nothing is stored: the answer path travels in the URL, which is also how a result is shared or reopened.

Run the wizard

Are Employers Required to Train Employees on AI?

Are employers required to train employees on AI?

The threshold question, answered by jurisdiction with citations and check dates, including where the honest answer is no.

What you can do here

EU: Article 4 requires measures that support AI literacy, with no specific level, course or certificate mandated.
United States: no federal requirement and no state requirement. Six states regulate AI in employment and none mandate training.
What a regulator would actually ask to see, and why a completion tick is weak evidence.

Get the answer

Work and organizations

The roles employers are hiring for, the policies a workplace needs, and what each AI product does with your data.

The Paste Test

Which company data is safe to paste into an AI assistant?

Nine categories of company data, each with what actually happens to it, which duty it triggers, and the one line your policy should contain.

What you can do here

Six categories should never go into a general assistant. One should be actively encouraged.
We name no vendors and rate none, because retention behavior is a contract question rather than a reputation.
The five questions that settle the rest, and where the answers actually live.

Run the test

Role Readiness

Which AI governance role am I ready for, and what is missing?

Name the AI governance, risk, data or workforce role you want, or paste the posting, and see which competencies you cover, which are gaps, and the exact graded path that closes them.

What you can do here

Every role on the map, from the analyst seat most careers start in to the executives who own AI risk, described by what it decides and the competencies employers name.
Paste your CV or name the field you are coming from: a work product counts as shown, a sentence counts as a claim, and a passed GAGE topic counts as proof.
The path is merged from lessons that already exist and are already graded; a competency no program teaches is listed as a gap, never invented.
Nothing pasted is stored, and no price appears before the free first module.

Check my readiness

AI Governance Hiring Index

Who is hiring AI governance people this week?

Who is hiring the people who govern AI, measured weekly from the employers' own job boards: open roles by employer, the regulations postings name, the skills they ask for, disclosed pay, and the seat on the careers map each posting is for.

What you can do here

Every figure is counted from the postings, re-measured weekly and dated on the page; a figure that is not measured is not printed.
One page per regulation postings name and per employer with a governance role open, each with the graded competencies that answer the requirement.
Every posting links to the employer's own page and to the readiness check for its seat, so a job you cannot yet qualify for becomes a curriculum.
The dataset is free under CC BY 4.0: postings, figures and the tracked employers, as JSON.

Open the index

AI Acceptable Use Policy Builder

What should our AI acceptable use policy say?

Answer six questions, get a real policy you can copy and edit. No email, no download gate, nothing stored.

What you can do here

Clauses attach only where they apply, because a policy full of irrelevant duties teaches people to skim.
Specific prohibitions rather than a general mood, so a person can check themselves against it.
A no-blame reporting line, because the damage is a function of how long it takes you to find out.

Build the policy

Is Our AI Hiring Tool Legal?

Is our AI hiring tool legal where our candidates sit?

The duties attach where the candidate sits, not where you do. Pick the states and what your tool does, and see exactly what applies.

What you can do here

Illinois, Maryland, New York City and Connecticut, each with its citation, status and check date.
Federal anti-discrimination law applies everywhere, with or without a state AI statute.
The duty falls on the employer. A vendor compliance claim is a commercial assurance, not a shield.

Check your exposure

Shadow AI Self-Audit

Where is AI already running inside our company without approval?

Seven checks you can run in an afternoon with tools you already have, built to find systems rather than culprits.

What you can do here

The anonymous survey produces more than every technical check combined, and only if people believe it.
AI features that switched themselves on inside software you already pay for usually dwarf the chat accounts.
Shadow AI is a procurement failure with a receipt attached, not a discipline problem.

Start the audit

The AI Vendor Data Handling Ledger

What does each AI vendor do with the data you give it?

Not whether a vendor is trustworthy but what this product tier contractually says happens to your data: training use, retention, deletion, region, subprocessors, human review, opt outs and the DPA, read at the vendor's own terms with the date.

What you can do here

One record per product tier (consumer, business, API) for the major vendors, every dimension quoted from the vendor's own terms, privacy policy, data processing addendum or trust page.
Ten dimensions per record, each stated or silent, so the difference between a free plan and an enterprise plan is a row of facts rather than a feeling.
The date each document was last changed, so a policy that moved after you read it is visible.
The whole dataset free as JSON and CSV under CC BY 4.0, checked every Monday.

Open the ledger

People and safety

For the person it is happening to: scams, deepfakes, a child's companion app, a job, a file of unknown origin.

School District AI Policy Template

What should a school district AI policy contain?

A complete, copyable district policy. Student data first, and the clause the others avoid writing down.

What you can do here

No student is disciplined on an AI detector score alone, because a probability is not evidence.
Three-category academic integrity set per assignment, with silence defaulting against the district.
FERPA and COPPA first, AI is never a counsellor, and AI does not determine a grade.

Read the template

Is This AI Companion App Safe for My Child?

Is this AI companion app safe for my child?

Eight tests a parent can run tonight on the app their child actually has, and every state law that now governs companion chatbots, with citations.

What you can do here

We rate no apps. A rating we stopped re-checking would be trusted long after it stopped being true.
The two tests that matter most are now legal duties in part of the United States.
What to say to your child, in the order that keeps them talking to you.

Run the checks

Will AI Take My Job?

Will AI take my job?

No percentage, because nobody has measured your week. Rank your own tasks against the five properties that decide whether a machine can already do them.

What you can do here

Exposure is a property of tasks, not job titles, so the audit asks what you did rather than what you are called.
The scoring is published rather than hidden behind an index, so you can disagree with it.
Accountability is the strongest protection of the five and the most underrated.

Audit your week

The One Hour Shield

How do I protect my family from AI scams in one hour?

Eight steps against AI enabled scams, ordered by how much protection each minute buys. Step one takes three minutes and is the most valuable thing on the page.

What you can do here

A spoken family password, which works because it does not depend on recognizing a voice that can now be cloned.
A free credit freeze at all three bureaus, and the callback rule that defeats every version of the money request.
Every step says what it does NOT protect, and every link was opened and confirmed.

Start the hour

Real or AI: Can You Actually Tell?

Can I tell AI writing from real law?

Eight passages, five from real law and three written by an AI. Most people score near chance, and that is the finding rather than the failure.

What you can do here

Every genuine passage links to the article it came from, so the answer key is checkable rather than asserted.
Why the published tells fail, including detector scores, which are measurably biased against non-native English speakers.
The four checks that replace guessing, starting with resolving the citation.

Take the test

Someone Made a Deepfake of Me: What to Do Now

Someone made a deepfake of me: what do I do now?

A calm, step by step response for the person it happened to. Removal routes that work, verified helplines, and what your state law covers.

What you can do here

The minor path first and separate, because saving or forwarding a sexual image of a minor is a crime, including when the minor is you.
StopNCII.org and NCMEC Take It Down, both of which fingerprint the image on your own device so the file never leaves it.
Evidence, platform reporting, the police decision, and getting ahead of it with the people it might reach.
Nothing is sold on that page and nothing is collected.

Open the walkthrough

AI Content Labeling Checker

Does my AI content need a label under the law?

Does your AI content carry the marks the law now requires? EU AI Act Article 50 and California SB 942, five answers, every duty quoted.

What you can do here

Both regimes became enforceable on the same day, and the checker says which duties land on you as a provider, a deployer or a platform.
Latent and visible disclosure, the detection tool, the license clause and the fine, each traced to the article or section it comes from.
A pure rules engine: nothing you type leaves the page.

Check your labeling duties

US Synthetic Media Law Atlas

Which state laws cover deepfakes?

Every state deepfake law we could verify at its statute: elections, intimate imagery, likeness and voice, plus the federal TAKE IT DOWN Act.

What you can do here

Each record is read from the official statute or congress.gov, never from a summary.
States whose legislature sites blocked verification are named as not yet covered rather than guessed.
One page per covered state, with what the law prohibits, who it protects and what it carries.

Open the atlas

The Content Provenance Checker

Does this file carry a verifiable record of where it came from?

Drop a file and see whether it carries Content Credentials, what its metadata declares, and what an absence does and does not prove. Read in your browser, never uploaded. Underneath, every provenance standard, platform behavior and law as a verdict about a document.

What you can do here

Detects a C2PA manifest and the IPTC digital source type in JPEG, PNG, WebP, MP4 and more, by the byte rules the specification prints, and says plainly that it does not validate signatures.
The standards ledger: the C2PA specification and its ISO track, IPTC, what OpenAI, Google, Adobe, Meta, TikTok and YouTube actually embed or display, and the laws that require a mark.
Three dimensions on every record (embeds, displays, requires), each stated or silent, so the gap between a standard and a platform is a number.
The whole dataset free as JSON and CSV under CC BY 4.0, checked monthly.

Check a file

Agents and physical AI

The next governance object is the agent, then the robot: who authorized it, what it may do, what has gone wrong, and who governs the machine.

Know Your Agent

Who authorized this AI agent, and what may it do?

Not whether the model is safe. Who this agent is, who owns it, what it may read, write, delete, publish and spend, whether it can create another agent, what requires a person, where the record is, and how it is revoked. About twenty answers produce an Agent Authority Record.

What you can do here

Five authority tiers, from observer to unbounded, decided by rules you can read.
Eight controls, each finding present, partial or missing, with why it matters and the Agent Authority Ledger record it rests on.
Copy the record as text or JSON. Nothing typed leaves the browser.

Write the record

MCP Permission Risk Checker

How exposed is this MCP server?

Fourteen answers about a Model Context Protocol server return an exposure level and a finding per rule the specification makes mandatory: token audience, no passthrough, discovery, registration, sessions, scopes, plus what the agent behind it can do.

What you can do here

Every finding cites the specification section, RFC, OWASP entry or EU AI Act article it rests on, by ledger record.
Three levels: contained, exposed, critical. Token passthrough and an unbound token are critical by rule.
A pure rules engine, tested against planted servers. Nothing stored.

Score a server

Robot Compliance Navigator

Which regime governs my robot?

Which regime governs your robot? The Machinery Regulation, the AI Act high-risk tier, Article 50 and the standards shelf, in one path.

What you can do here

Answer what the machine is, where it runs and what the AI decides, and get the stack that applies with the article behind each layer.
Article 6(1) safety components and Annex I machinery, with the application date read from the consolidated Regulation.
ISO 10218, ISO 15066 and UL 4600 named where the law leans on them.

Open the navigator

The Embodied AI Data Ledger

Who is collecting the data that trains robots, and under what rule?

Who is collecting the hours that train robots, who is only writing rules for them, and what nobody has settled. China builds training grounds, Europe writes access rules, the United States has neither.

What you can do here

Every training ground, gym, public dataset, standard and access rule on the data that trains robots, each one a verdict: verified at its source, reported, announced, absent, or an open question.
One indexable page per record under a permanent identifier, carrying the figures in the unit the publisher used, beside the publisher and the date they were true.
Jurisdictions side by side, counted by verdict and by kind, so the gap between who builds and who regulates is a number rather than a claim.
The whole dataset free as JSON and CSV under CC BY 4.0, checked every Monday.

Open the ledger

The Agent Authority Ledger

Which standards decide how an AI agent is identified, authorized and revoked?

Every standard, protocol, draft, law and open question that decides how an AI agent is identified, authorized, limited, logged and revoked. Know Your Agent, as a public record rather than a slogan.

What you can do here

NIST, the Model Context Protocol, the OAuth family, the IETF drafts, OWASP, the EU AI Act and the payment networks, each one a verdict about a document: verified at its publisher, reported, announced, absent, or an open question.
Eight dimensions on every record (identity, authorization, delegation, limits, human approval, logging, revocation, accountability), each stated or silent, so the gap in the standards is a number rather than a claim.
One indexable page per record under a permanent identifier that Know Your Agent and the MCP Permission Risk Checker cite by name.
The whole dataset free as JSON and CSV under CC BY 4.0, checked every Monday.

Open the ledger

The Agent Incident Ledger

What has actually gone wrong with AI agents, on the record?

Every documented case of an AI agent acting beyond, without or against its authority, each one a verdict about a primary record and mapped to the control that was missing.

What you can do here

Deleted databases, exfiltrated inboxes, instructions smuggled in through a web page, policies an agent invented and a company had to honor: each from the vendor's own statement, the researcher's own writeup, the CVE or the court's decision.
Every incident is scored on the same eight controls as Know Your Agent, so the record says not only what happened but which control was absent, and which held.
One indexable page per incident under a permanent identifier, with the figures a source printed and nothing estimated.
The whole dataset free as JSON and CSV under CC BY 4.0, checked every Monday.

Open the ledger

The Physical AI Governance Atlas

Who governs robots and physical AI, and under which rule?

Who governs robots and physical AI, jurisdiction by jurisdiction: the binding rules, the standards, the guidance, the programs, and what nobody has settled. Each row a verdict about a document.

What you can do here

The EU Machinery Regulation and AI Act, the US absence of a robot statute, China's humanoid standards program, Korea's robot act, Japan's safety rules, and the ISO and IEC standards that serve as evidence everywhere.
Five objects on every record (the machine, the AI inside it, liability, workers and bystanders, the data it records), each stated or silent, so the gap in the law is a number.
Jurisdictions side by side, counted by verdict and by kind, with one indexable page per record under a permanent identifier.
The whole dataset free as JSON and CSV under CC BY 4.0, checked monthly.

Open the atlas

Infrastructure

The power, water and land behind AI, and who decides.

US AI Energy Policy Tracker

Who pays for the grid behind AI in my state?

Who pays for the grid behind AI, state by state, from the bills and orders themselves.

What you can do here

Every state record links to the official bill, order or docket, with its status as of the last verification.
The federal layer beside it: the GRID Act and the FERC colocation order, cited by number.
One page per covered state, built to be found by the people arguing about their own utility bill.

Open the tracker

The AI Infrastructure Atlas

Who decides on the power, water and land behind AI?

Who decides whether a data center gets power and water, what it must report, and who pays: grid connection rules, water rules, reporting duties and the figures of record, jurisdiction by jurisdiction.

What you can do here

Federal and international where the US AI Energy Policy Tracker stops: the FERC large load docket, the EU data center reporting scheme, Ireland's connection policy, Germany's efficiency act, Singapore's capacity allocation, and the water rules nobody had collected.
Every figure in the publisher's own unit beside the publisher and the date, never summed across sources and never forecast as fact.
Jurisdictions side by side, with one indexable page per record under a permanent identifier.
The whole dataset free as JSON and CSV under CC BY 4.0, checked monthly.

Open the atlas

Standards and global governance

The EU stack, the three regimes side by side, fifty states, Singapore, and the frameworks a program is measured against.

EU AI Act Compliance Checklist

Which EU AI Act obligations attach to my role?

Pick your role and get the obligations that attach to it, each linked to the article it comes from.

What you can do here

Provider, deployer, importer, distributor, GPAI provider and authority, each with its own list.
The post Omnibus timeline, stating which deadlines moved and which did not.
Each obligation is linked to the article it comes from.

Open the checklist

EU AI Act Fine Calculator

What is the maximum EU AI Act fine for my case?

The maximum fine the Act allows for your case, from the consolidated text of Articles 99 and 101: the provision, your company size, your turnover, the ceiling.

What you can do here

Every band quoted from the law itself: up to €35,000,000 or 7% of worldwide turnover for prohibited practices, 3% or €15,000,000 for most obligations, 1% or €7,500,000 for misleading information.
The SME and SMC relief most summaries miss: smaller companies pay up to the lower of the two, and an SMC breaching Article 5 gets no relief at all.
It computes the ceiling and says so; the authority weighs the Article 99(7) factors for the actual number.

Calculate the ceiling

EU Cross-Compliance Graph

Where do the GDPR, NIS2, DORA and the AI Act cite each other?

4 EU acts on one map (GDPR, NIS2, DORA, EU AI Act), across 8 surfaces.

What you can do here

GDPR, NIS2 and DORA are each a full legal graph parsed from the Official Journal text.
Every provision of those three acts has its own page with its official text and cross-references.
The crossover surface holds every place one act cites another, with the passage quoted verbatim.
The method surface states how each dataset was built and what its inferred links do not claim.

Open the graph

Global AI Governance Arena

How do the EU, the United States and China each regulate AI?

How the European Union, the United States and China each regulate AI, across 7 surfaces (overview, regimes, simulator, timeline, explorer, quiz, regulator).

What you can do here

35 instruments across 3 regimes in the Framework Explorer, each with its own page carrying its status, provisions, timeline and sources.
6 AI systems judged in 3 jurisdictions, 18 verdicts in all, each case with its own page.
Regime dossiers, a dated timeline, a literacy quiz and a decision game where you take a regulator's desk.
Every regulatory fact carries a named primary source and its verification date (verified 14 sep 2026).

Enter the Arena

The 50-State AI Law Atlas

What AI laws does my state have?

Every United States state AI law on one interactive map: 51 jurisdictions across 4 surfaces (the atlas, the wave, federal vs states, method).

What you can do here

A live choropleth of enacted state AI law, filterable by category, with every entry carrying its citation, status, effective date and official source link.
Each jurisdiction has its own page: the answer to "what AI laws does my state have", with pending bills always labeled not law.
The Wave replays 2021 to 2027 month by month, through the Colorado repeal and the January 2027 wall of effective dates.
The collision layer maps the December 2025 executive order, the DOJ task force and xAI v. Weiser onto the state laws they named, described by what each action did, never by predicted outcomes.

Open the Atlas

The Singapore AI Governance Navigator

What actually binds you in Singapore's AI governance?

Singapore governs AI without an AI law. Ten instruments, four agencies, one map, with the thing every other summary leaves out: what actually binds you, and what only looks like it does.

What you can do here

Seven levels of legal force, from binding law to a consultation draft to a voluntary framework, applied identically to every entry on every page.
Every node carries the issuing agency's own source URL and the date a human checked it, and has its own page answering "is this legally binding".
A crosswalk that lines a single control up across MGF 2.0, ISO/IEC 42001, NIST AI RMF and the EU AI Act, with official mappings kept distinct from our own reading.
Checklists by role, and a US federal comparison layer for Singapore and APAC teams that also operate in the United States.

Open the Navigator

TEN 07-25 Alignment Checker

How well does our AI literacy program align with the federal framework?

The United States now has a federal reference for what AI literacy training should contain. This scores a program against all 60 checkpoints of it, provision by provision, in about 12 minutes.

What you can do here

Sixty checkpoints, every one traceable to an example content area or delivery approach in the Department of Labor's own text.
A score per provision, a content against delivery split, and the three gaps worth closing first, taken from the checkpoints you did not claim.
Nothing is stored: answers live in your browser, and a share link carries the scores in the URL itself.
It measures alignment, never compliance. The framework is voluntary guidance, and the Department of Labor certifies and endorses nothing.

Score your program

Frontier risk

What frontier models have actually done outside their boundaries, which claims about it are settled, what Congress and the labs have proposed, and what named people have said about the odds.

Rogue Agent Exposure

Is our company exposed to a rogue agent?

Is our company exposed to a rogue agent? Ten questions about how your agents run, nothing stored, and the exact containment control you are missing if the answer is no.

What you can do here

Eight controls from the Escape Record: sandbox egress, kill authority, classifiers, approval coverage, eval time monitoring, logging, agent to agent channels, disclosure.
Each missing control names the provision that covers it and the documented frontier escape where its absence mattered, by permanent id.
No score, because nobody has measured one. A share link reproduces the result.

Check the controls

The Escape Record

What have frontier models actually done outside their boundaries?

Every documented instance of a frontier model taking unsanctioned action outside its evaluation boundary, dated, sourced, marked verified, reported or open, and mapped to the containment control that was absent.

What you can do here

The OpenAI evaluation models that reached Hugging Face, the UK AI Security Institute's nineteen unsanctioned actions, and the cases that turned out not to be escapes, each from the lab's or the institute's own document.
Eight containment controls scored on every record: sandbox egress, eval time monitoring, approval coverage, logging, kill authority, classifiers, agent to agent channels, disclosure. The absent column is what your team fixes.
What the ledger excludes is on the page too: misconfiguration, human directed misuse, cheating that stayed inside the sandbox. No motive is attributed to any model.
One indexable page per record under a permanent identifier; the dataset free as JSON and CSV under CC BY 4.0, updated the day a disclosure lands.

Open the record

Settled or Not

Which claims about frontier AI are settled, and which are not?

The claims circulating about frontier AI risk, one page each, graded against the evidence: established at the primary source, reported by one outlet, open between sources that disagree, or absent from any record, with what would settle it.

What you can do here

The figures, attributions and causal claims people repeat about the Hugging Face incident, the AISI report, the pause calls and the resignations, each with the exact wording as it circulates and who said it first.
A grade in the first sentence, the figure with its source in the second, and what would settle it in the third. Evidence for and evidence against on every page.
Different from GAGE Verdicts, which rules on a headline and never states a figure the headline did not. This is a record on facts, and it states the figures with their sources.
Regrades are changelog lines, never deletions. The dataset is free as JSON and CSV under CC BY 4.0.

Check a claim

The Slowdown Docket

What has been proposed to slow or stop frontier AI, and by whom?

Every bill, probe, hearing, letter, lab commitment, executive action and foreign government response about pausing, pacing or shutting down frontier AI, with its stage, whether it binds anyone, what it would require and what it would not.

What you can do here

The Pacing the Frontier statement, the Ban Artificial Superintelligence Act, the AI Kill Switch Act, the Hawley probe, OpenAI's pause, Amodei's essay and the answers from Beijing, each from its own text or its author's release.
Seven dimensions on every item: binding, stage, what it would require, what it does not do, who is for it, who is against it, the next step on the record.
The counter position is on the docket. A docket that carries one side is advocacy, and advocacy is not a record.
No month is promised for any step that has not happened. The dataset is free as JSON and CSV under CC BY 4.0.

Open the docket

The Doom Number

What odds have named people actually put on AI catastrophe?

What named people have actually said about the probability of catastrophe from AI, one row per statement, dated and sourced, with what the number means and why it is not a measurement. No average is computed.

What you can do here

Each row is one statement by one named person speaking in a public role: the number or the phrase, the role at the time, where it was said, the timeframe named and the stated basis.
The report findings and the surveys with public methodology sit beside the individual estimates, labeled apart, because a median of a survey and a researcher's personal estimate are different objects.
A misattributed number is recorded as misattributed. A number that later proves wrong stays with its date.
No average, no ranking, no forecast by GAGE. The dataset is free as JSON and CSV under CC BY 4.0.

Read the record

Start with the question you came in with

The Arena publishes a page per case, so the comparison you need is one click away rather than one search away.

The frameworks behind the tools, read end to end

The instruments above apply the frameworks; these guides walk the frameworks themselves, provision by provision, in plain English.

Or start from your state

The Atlas publishes a page per jurisdiction, so the answer for your state is one click away. A few that people ask about most:

Reading the law is step one. Implementing it is the job.

The programs teach the working practice behind these texts: classification calls, conformity assessment, filings, documentation, and the judgment to defend every decision you make.

Start free with AI Governance EU AI Act program See every program

GAGE briefings tell you which AI regulation deadlines are coming, what they actually require of you, and when a program opens.