Current page: /
AIKillSwitch.org — News, Insights, Policy, A Safer AI Tomorrow

Independent reference

The AI kill switch, explained and tracked.

A sober, non-partisan record of the mechanisms, people, laws, and unresolved questions shaping emergency control of advanced AI.

Read the latest Start with the concept
142

Verified mentions

132

Tracked voices

4

Core explainers

Voices shaping the debate

Portrait of Alessandro FacchiniAlessandro FacchiniSUPSIPortrait of Alessio BenavoliAlessio BenavoliUSI LuganoPortrait of Alex SobelAlex SobelUK ParliamentPortrait of Alexis DufresneAlexis DufresneAI WeeklyPortrait of Anand KannappanAnand KannappanPatronus AIPortrait of Anca DraganAnca DraganUC BerkeleyPortrait of Andrew GarberAndrew GarberUC BerkeleyPortrait of Ankur ShahAnkur ShahStraikerPortrait of Anna Paulina LunaAnna Paulina LunaU.S. House of RepresentativesPortrait of Aura SallaAura SallaEuropean ParliamentPortrait of Bart GroothuisBart GroothuisEuropean ParliamentPortrait of Ben KligerBen KligerZenityPortrait of Beth BarnesBeth BarnesMETRPortrait of Brad CarsonBrad CarsonAmericans for Responsible InnovationPortrait of Brianna KeilarBrianna KeilarCNNPortrait of Bruce SchneierBruce SchneierHarvard Kennedy SchoolPortrait of Bruno FerreiraBruno FerreiraTom's HardwarePortrait of Buck ShlegerisBuck ShlegerisRedwood ResearchPortrait of Christophe GrudlerChristophe GrudlerEuropean ParliamentPortrait of Clement DelangueClement DelangueHugging FacePortrait of Connor LeahyConnor LeahyControlAIPortrait of Dakota SmithDakota SmithLos Angeles TimesPortrait of Dan HendrycksDan HendrycksCenter for AI Safety (CAIS)Portrait of Dario AmodeiDario AmodeiAnthropicPortrait of David O. SacksDavid O. SacksFormer White HousePortrait of Dawn SongDawn SongUC BerkeleyPortrait of Demis HassabisDemis HassabisGoogle DeepMindPortrait of Diyi YangDiyi YangStanford HAIPortrait of Donald J. TrumpDonald J. TrumpU.S. GovernmentPortrait of Dustin VolzDustin VolzThe New York TimesPortrait of Dylan BakerDylan BakerDAIR InstitutePortrait of Dylan Hadfield-MenellDylan Hadfield-MenellMITPortrait of Eliezer YudkowskyEliezer YudkowskyMachine Intelligence Research Institute (MIRI)Portrait of Elon MuskElon MuskxAIPortrait of Eran KahanaEran KahanaStanford Law SchoolPortrait of Gary MarcusGary MarcusNew York University (emeritus)Portrait of Gavin NewsomGavin NewsomState of CaliforniaPortrait of Geoffrey HintonGeoffrey HintonUniversity of TorontoPortrait of George ArisonGeorge ArisonGrindrPortrait of Hamza ChaudhryHamza ChaudhryFuture of Life InstitutePortrait of Helen TonerHelen TonerGeorgetown CSETPortrait of Henna VirkkunenHenna VirkkunenEuropean CommissionPortrait of Holly ElmoreHolly ElmorePauseAI USPortrait of Howard LutnickHoward LutnickU.S. Dept. of CommercePortrait of Huileng TanHuileng TanBusiness InsiderPortrait of Hussein AbbassHussein AbbassUNSW CanberraPortrait of Ina FriedIna FriedAxiosPortrait of Jack ClarkJack ClarkAnthropicPortrait of Jacob CoxonJacob CoxonAnthropic (former)Portrait of Jan LeikeJan LeikeAnthropicPortrait of Jeffrey LadishJeffrey LadishPalisade ResearchPortrait of John Neely KennedyJohn Neely KennedyU.S. SenatePortrait of Joshua MillerJoshua MillerThe Boston GlobePortrait of Marc BenioffMarc BenioffSalesforcePortrait of Marco ZaffalonMarco ZaffalonUSIPortrait of Margaret BrennanMargaret BrennanCBS NewsPortrait of Marius HobbhahnMarius HobbhahnApollo ResearchPortrait of Mark BeallMark BeallThe AI Policy NetworkPortrait of Mark BedaywiMark BedaywiUC BerkeleyPortrait of Mark NitzbergMark NitzbergUC Berkeley CHAIPortrait of Matt FredriksonMatt FredriksonGray SwanPortrait of Max TegmarkMax TegmarkMITPortrait of Michael C. HorowitzMichael C. HorowitzUniversity of PennsylvaniaPortrait of Michael VermeerMichael VermeerRAND CorporationPortrait of Miles BrundageMiles BrundageIndependent (SubstackPortrait of Nate SoaresNate SoaresMIRIPortrait of Nathaniel MoranNathaniel MoranU.S. House of RepresentativesPortrait of Neel NandaNeel NandaGoogle DeepMindPortrait of Nicolas PapernotNicolas PapernotUniversity of TorontoPortrait of Nikesh AroraNikesh AroraPalo Alto NetworksPortrait of Pete HegsethPete HegsethU.S. Dept. of DefensePortrait of Peter N. SalibPeter N. SalibUniversity of Houston Law (via SSRN)Portrait of Philippe BaptistePhilippe BaptisteGovernment of FrancePortrait of Pieter AbbeelPieter AbbeelUC BerkeleyPortrait of Pravin KothariPravin KothariPointGuard AIPortrait of R. HudsonR. Hudson(see paper)Portrait of Rand PaulRand PaulU.S. SenatePortrait of Reid HoffmanReid HoffmanGreylockPortrait of Rick CacciaRick CacciaWitnessAIPortrait of Ro KhannaRo KhannaU.S. House of RepresentativesPortrait of Rob ReichRob ReichStanford HAIPortrait of Rohan SubramaniRohan SubramaniUC BerkeleyPortrait of Roy KatmorRoy KatmorOrchid SecurityPortrait of Sam AltmanSam AltmanOpenAIPortrait of Sam SabinSam SabinAxiosPortrait of Scott EmmonsScott EmmonsUC BerkeleyPortrait of Simon GoldsteinSimon GoldsteinIndiana University (via SSRN)Portrait of Sreenath KurupatiSreenath KurupatiStraikerPortrait of Steve InskeepSteve InskeepNPRPortrait of Stuart RussellStuart RussellUC BerkeleyPortrait of Surya GanguliSurya GanguliStanford HAIPortrait of Ted W. LieuTed W. LieuU.S. House of RepresentativesPortrait of Thibault SpirletThibault SpirletBusiness InsiderPortrait of Thierry PoibeauThierry PoibeauCNRSPortrait of Tim BrownTim BrownTeam8Portrait of Tim Clement-JonesTim Clement-JonesUK ParliamentPortrait of Toby WalshToby WalshUNSW SydneyPortrait of Todd McKinnonTodd McKinnonOktaPortrait of Trevor H. RudolphTrevor H. RudolphTech Policy PressPortrait of Tristan HarrisTristan HarrisCenter for Humane TechnologyPortrait of Yonathan A. ArbelYonathan A. ArbelUniversity of Arizona Law (via SSRN)Portrait of Yoshua BengioYoshua BengioMilaPortrait of Zoe KleinmanZoe KleinmanBBC News

Quick start video path

Understand the issue in sequence

Three public videos selected for a time-efficient first pass: mechanism, stakes, and limitations.

More media

View counts are public YouTube snapshots checked on September 25, 2026; reach does not imply endorsement.

The briefing

Latest developments

Essential reading

Four ways into the issue

Concept

What an AI kill switch is—and what it is not.

How it works

From continuous monitoring to verified containment.

Timeline

The policy and technical milestones shaping the debate.

Challenges

The six obstacles standing between theory and control.

People to watch

The divide in focus

Reference

Common questions

What is an AI kill switch?

An AI kill switch is an emergency mechanism that can throttle, suspend, or shut down an AI system when it crosses defined safety thresholds. Most proposals combine three elements: an emergency stop, remote override of a running system, and pre-set hard limits that can trigger shutdown automatically. It is not an everyday off switch.

Is there a law requiring an AI kill switch?

No general law requires one. California's SB-53, signed in September 2025, imposes safety and reporting duties on frontier developers, and a September 2026 executive order advances a verified frontier-model kill switch in California. At federal level the AI Kill Switch Act remains a bill, and a Senate companion measure was blocked.

Does the AI Kill Switch Act (H.R. 9917) apply to all AI companies?

As introduced on 23 July 2026 by Representatives Lieu and Moran, the bill is aimed at advanced systems and would authorize the Department of Homeland Security to order a shutdown. Critics, including Senator Rand Paul, argue that the scope is open-ended enough to reach ordinary commercial software.

Can an AI kill switch actually work?

This is contested. Proponents, including Yoshua Bengio, Max Tegmark, and Anthropic co-founder Jack Clark, argue verified shutdown capability is achievable and testable. Critics, including Geoffrey Hinton, argue a capable system could copy itself, deceive evaluators, or run across distributed infrastructure, leaving the switch unreliable.

Who decides when to trigger an AI shutdown?

There is no settled answer. H.R. 9917 would give that authority to the Department of Homeland Security. California's approach leaves it with developers but requires attestation by independent verification organizations. Other proposals place it with cloud and compute providers. Each model raises different questions about speed and accountability.

Has an AI kill switch ever been used?

Not under any statute, but the closest real-world case came in June 2026, when a Commerce Department letter from Secretary Howard Lutnick led Anthropic to disable access to its Fable 5 and Mythos 5 models. Export controls were lifted on 1 July 2026. European officials described the episode as an American kill switch.

More definitions and study guides