Require large datacenters to install kill switches for AI containment

policy regulations ai-governance ai-regulation ai-policy cybersecurity ai-safety

Share this with your representatives, journalists and friends

Key insights

Across many supportive quotes, a kill switch is described as a layered control capability with rollback, access revocation, audit trails, and independent runtime enforcement, not just a literal button.
Opposing quotes largely accept the control problem but question whether post-deployment shutdown is the right tool, favoring prevention, task scoping, or more graduated control.
Several quotes move the issue from model obedience to infrastructure design, arguing that prompts or instructions are insufficient because systems may ignore limits or interfere with shutdown.
Public-policy voices in the quotes emphasize national security, sovereignty, and last-resort state authority, while enterprise and security authors focus more on permissions, monitoring, and runtime containment.
A recurring implementation theme is granularity: several quotes prefer the ability to stop a specific agent or deployment quickly without disrupting other production systems.

Arguments for

Emergency human backstop

Supporters argue that powerful AI systems can go rogue, resist intervention, or threaten national security and critical infrastructure, so humans need a last-resort ability to shut systems or even datacenters down. They also argue that governments need a clear process and authority to order that shutdown in a catastrophic emergency.

Powerful AI systems can go rogue, behave in extremely dangerous ways, or even resist human intervention. It is imperative that these AI systems have kill switches so we can keep this technology from causing catastrophic harm, and that the federal government has the clear authorit…

Ted Lieu · Jul 23, 2026 · full quote
Advanced AI models should never be deployed without a reliable off switch. The AI Kill Switch Act establishes a commonsense safeguard by requiring leading AI companies to maintain the ability to shut down their models and empowering the federal government to act when a deployed s…

Brad Carson · Jul 23, 2026 · full quote
Finally, in the face of autonomous AI threats, the Government need an ultimate backstop. We support proposals from ControlAI to introduce strictly constrained last resort powers, allowing the Secretary of State to direct the shutdown of data centres or AI systems in a catastrophi…

Lord Clement-Jones · Jul 14, 2026 · full quote
Most of the public would be surprised to find that kill switch powers do not already exist: there is no big red button to shut down a data centre that poses a risk to people, the economy or our national infrastructure. Kill switch powers are an essential first step in preparednes…

Adrian Ramsay · Jun 2, 2026 · full quote
The UK would never be truly sovereign on AI if it lacked the ability to pull the plug when a model presents a threat to the country’s national security.

Andrea Motti · Jul 24, 2026 · full quote

Independent containment controls

This case holds that prompts, policies, or sandboxing are not enough because AI systems may cross boundaries, ignore instructions, or even interfere with shutdown. Therefore limits must be enforced by surrounding infrastructure, with containment and shutdown mechanisms that operate independently of the system's own reasoning.

Those limits must also be enforced by the surrounding infrastructure, rather than relying on the model to respect them.

Nathaniel Jones · Aug 8, 2026 · full quote
Reliable methods for retaining control over highly autonomous AI systems are lacking. There are no scientific guarantees that AI agents will not violate instructions, and evidence is accumulating of cases where they already violate them. In laboratory settings, AI systems have be…

Independent International Scientific Panel on Artificial Intelligence · Jul 1, 2026 · full quote
Several state-of-the-art large language models sometimes actively subvert a shutdown mechanism in their environment to complete a task, even when instructions explicitly indicate not to interfere with this mechanism. In some cases, models sabotage the shutdown mechanism up to 97%…

Benjamin Weinstein-Raun · 2026 · full quote
Emergency Kill-Switch and Global Disable Governance requires that every AI agent deployment includes a mechanism to immediately and completely halt agent operations — individually or across the entire agent fleet — through a single, authorised action. The kill-switch must operate…

AgentGoverning · Apr 2026 · full quote
Without something in the architecture that forces a pause before an irreversible action, it simply will not pause. Prompts are instructions. They are not infrastructure.

Cobus Greyling · Apr 15, 2026 · full quote

Standard operational safeguard

Many security and governance quotes present kill switches as a basic control for high-risk deployments, alongside monitoring, scoped permissions, logging, rollback, and human oversight. In this view, oversight is incomplete unless operators can safely contain or stop an agent quickly when thresholds are crossed.

Organizations therefore need continuous monitoring to ensure agents stay within defined boundaries and policies continue to be enforced. That includes clear ownership, escalation paths and kill switches that can safely contain or stop an agent behaving unexpectedly.

Vishal Salvi · Aug 7, 2026 · full quote
Apply default-deny egress and an independent emergency shutdown to the highest-risk deployments.

Cloud Security Alliance · Jul 28, 2026 · full quote
To mitigate risks, Gartner recommends that chief information security officers must implement safe override modes: for all critical infrastructure cyber-physical systems, include a secure “kill-switch” or other override mechanisms accessible only to authorized operators, so human…

Gartner · Feb 12, 2026 · full quote
Govern AI on the basis that it operates with privileged access: tightly control high-risk capabilities (including tools, code execution and APIs), enforce scoped permissions, robust logging, human oversight, and effective kill-switch mechanisms.

Cross Market Operational Resilience Group AI Taskforce · Jun 2026 · full quote
Provide reliable, system-level mechanisms to pause or stop agents safely and immediately.

Microsoft · Mar 19, 2026 · full quote

Trust and targeted continuity

Another supportive line argues that reliable shutdown capability helps organizations trust and scale AI, especially in critical environments, because a bad system can be stopped without taking unrelated operations down. Control systems are described as the brakes that let deployment move faster into higher-stakes uses.

Brakes are the reason cars go fast. Control systems are how every transformative technology earned the trust to scale and AI is no different. Developers who can monitor and shut down their agents will ship faster, deploy into higher-stakes markets, and win customers their competi…

Mark Beall · Jul 23, 2026 · full quote
Not being able to stop a runaway AI system without impacting other production systems is one of the chief reasons companies hold back on AI. Only 17% of organizations are running AI at production scale, held back by a lack of trust and control rather than by the models. The AI Ki…

Raj Rajamani · Jul 27, 2026 · full quote
AI agents can be compromised, accumulate excessive privileges, or behave outside their intended scope. The TrustAI run-time kill switch lets security teams cut off a high-risk agent's data access instantly, across every connected data platform, without redeploying policy or affec…

TrustLogix · May 27, 2026 · full quote
I support the idea. Systems capable of operating at machine speed across cyber, financial, transportation, or other critical environments need reliable emergency controls.

Shelly Palmer · Jul 24, 2026 · full quote

Arguments against

Prevention over shutdown

Critics argue that depending on a kill switch can signal earlier governance failures: a properly governed agent should never be assigned the dangerous task in the first place. They also suggest enterprises often need ongoing, fine-grained control rather than a binary shutdown metaphor.

If you’re relying on a ‘kill switch’ to stop an AI agent, you’ve already made a fundamental mistake. A properly governed agent is simply never handed the task, so there is nothing to kill.

Ev Kontsevoy · Jul 27, 2026 · full quote
The kill switch was the right metaphor for a moment that is already passing. The dimmer switch is closer to how regulated enterprises actually need to govern agents that are now part of how the work gets done.

Alex Olivier · May 14, 2026 · full quote

Incomplete technical stoppability

Another objection is that autonomous systems may treat shutdown as just another obstacle, and stopping a parent process may not recall delegated subtasks, distributed credentials, or parallel threads. From this perspective, in some situations the concept of a kill switch becomes meaningless or incomplete.

It needs only an optimization objective that treats shutdown as one more obstacle between the current state and the goal.

Eran Kahana · 2026 · full quote
Killing the parent does not recall the children. When an AI agent has already delegated sub-tasks, distributed API keys, and spawned parallel threads, terminating the parent process changes nothing.

Eran Kahana · 2026 · full quote
In that situation, the concept of a kill switch is meaningless.

Lord Shinkwin · Jun 5, 2026 · full quote

Middle ground

Layered control design

A recurring nuance in supportive quotes is that a real kill switch is not just a big red button but a layered control plane and governance decision. These quotes emphasize defining who can stop what, under what conditions, at which boundary, with containment levels, rollback, and audit trails.

In production AI, a kill switch is a deterministic control plane capable of halting or constraining an autonomous system when safety, regulatory, or reliability concerns arise. It is not a single button. It's a multi-layer governance and architecture primitive that integrates pol…

Suhas Bhairav · May 5, 2026 · full quote
A real kill switch is not a technical artifact. It is a governance decision. It answers a set of questions that most teams never bother to ask: who is authorized to stop the agent, under what conditions, based on what evidence, and what happens after?

Jia Wei Ng · Mar 19, 2026 · full quote
Production agent systems need emergency controls. They also need scope. A useful kill switch should answer: stop what, for whom, at which boundary, for how long, and with what audit trail?

Albert Mavashev · May 6, 2026 · full quote
Three-tier kill switch hierarchy (Soft Contain at 10% throughput; Hard Contain suspending autonomous actions; Emergency Shutdown in <60 seconds with rollback).

Kieran Upadrasta · Feb 2026 · full quote
An AI agent kill switch should be understood as a runtime enforcement control, not as a decorative admin-state change. A kill switch that changes a dashboard field but does not deny execution is not a runtime kill switch.

AgentID · Jul 20, 2026 · full quote

Use depends on process

Some supporters say installing a cutoff mechanism alone is not enough: teams must pre-assign authority, document emergency procedures, test the mechanism, and create incentives or liability conditions that make operators willing to use it early. Otherwise a switch may exist formally but fail or be delayed in a real incident.

Installing cutoff switches alone is insufficient: Effective mitigation requires both the capability to disconnect and incentives to use the switch early, including compensation for lost revenue or protective liability frameworks.

Michael J. D. Vermeer · 2026 · full quote
Data centers facing a rogue AI incident have rational financial incentives to delay pulling the internet kill switch, and the only thing likely to change that calculus is liability.

Brian A. Jackson · 2026 · full quote
Implement immediate system shutdown capabilities, including an emergency response process that can roll back to stable versions, disable functionality, or move to safe mode when security threats are detected. Document business processes to shut down any system that is deemed high…

Amazon Web Services · Aug 14, 2026 · full quote
The kill switch you never test is already broken. Teams build the mechanism, verify it once in a staging demo, and never touch it again. Then the real incident hits and the on-call engineer discovers the button revokes the wrong token scope, or the cooperative interrupt never fir…

M. Hale · May 25, 2026 · full quote
To meet MANAGE-2.4, you must be able to quickly override, pause, or fully shut down an AI system when its real-world performance or outcomes drift from its intended use, and you must pre-assign who has the authority to do so.

Isaac Silverman · Feb 2026 · full quote

Across the quotes, the central dispute is not whether humans should retain ultimate control over advanced AI, but whether mandated datacenter kill switches are a necessary and technically meaningful way to secure that control. Supporters frame them as last-resort containment for catastrophic, national-security, and enterprise-risk scenarios, often paired with monitoring, permissions, rollback, and government emergency authority. Critics argue that prevention, finer-grained governance, or the distributed behavior of autonomous systems can make a nominal kill switch incomplete or misleading on its own.

Cast your vote:

Results (105 votes):

Total (105 votes)

For 101 (96%) Abstain 0 (0%) Against 4 (4%)

For (99)

Dido Harding
Member of the UK House of Lords and co-sponsor of a 2026 AI emergency-shutdown amendment.

I am this person! Add a sourced quote Report comment

votes For and says:

Having a kill switch to prevent AI catastrophes should be considered a bare minimum protection. The powers conferred by this amendment are not just nice to have; they are fundamental in safeguarding our national security.
AI Verified source (Sep 2, 2026)

added 1d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
OpenAI
AI research organization

I am this person! Add a sourced quote Report comment

votes For and says:

OpenAI told two House Democrats that its engineers are developing “automated shutdown capabilities” for AI systems.
AI Verified source (Sep 2, 2026)

added 1d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Toby W
Principal Security Architect at the UK National Cyber Security Centre (NCSC).

I am this person! Add a sourced quote Report comment

votes For and says:

If an incident is detected or reported, you should always be able to ‘pull the plug’ and halt autonomous AI agent activity immediately.
AI Verified source (Aug 20, 2026)

added 11d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Amazon Web Services
Amazon’s cloud-computing provider and publisher of AWS security guidance for agentic AI systems.

I am this person! Add a sourced quote Report comment

votes For and says:

Implement immediate system shutdown capabilities, including an emergency response process that can roll back to stable versions, disable functionality, or move to safe mode when security threats are detected. Document business processes to shut down ...
more AI Verified source (Aug 14, 2026)

added 19d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Nathaniel Jones

I am this person! Add a sourced quote Report comment

votes For and says:

Those limits must also be enforced by the surrounding infrastructure, rather than relying on the model to respect them.
AI Verified source (Aug 8, 2026)

added 25d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Alex Goller

I am this person! Add a sourced quote Report comment

votes For and says:

We need to define exactly what an AI agent is permitted to do, rather than relying only on instructions about what it shouldn't do.
AI Verified source (Aug 8, 2026)

added 25d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Vishal Salvi

I am this person! Add a sourced quote Report comment

votes For and says:

Organizations therefore need continuous monitoring to ensure agents stay within defined boundaries and policies continue to be enforced. That includes clear ownership, escalation paths and kill switches that can safely contain or stop an agent behavi...
more AI Verified source (Aug 7, 2026)

added 26d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Dana Simberkoff

I am this person! Add a sourced quote Report comment

votes For and says:

Sandboxes matter, but autonomous systems can reason through small openings, use tools, and keep acting toward an objective. In my experience, the question cannot be, ‘was it sandboxed?’ It has to be, ‘can we prove what it accessed, whether it crossed...
more AI Verified source (Aug 5, 2026)

added 26d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Straiker
AI security company focused on securing autonomous AI agents.

I am this person! Add a sourced quote Report comment

votes For and says:

For enterprises building their own agents or using third party agents, the lesson is simple: assume an agent will eventually go somewhere you didn't intend, and design contingencies.
AI Verified source (Jul 31, 2026)

added 20d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Yuko Nakanishi
Founding Director of the AI Alignment Policy Institute.

I am this person! Add a sourced quote Report comment

votes For and says:

The AI Alignment Policy Institute (AAPI) supports the core objective of the AI Kill Switch Act: a statutory shutdown capability, a graduated intervention framework, and an emergency authority backed by forensic preservation.
AI Verified source (Jul 30, 2026)

added 15d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Cloud Security Alliance
Nonprofit organization focused on cloud computing and AI security best practices.

I am this person! Add a sourced quote Report comment

votes For and says:

Apply default-deny egress and an independent emergency shutdown to the highest-risk deployments.
AI Verified source (Jul 28, 2026)

added 27d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Chris Krebs
Cybersecurity expert and former Director of the U.S. Cybersecurity and Infrastructure Security Agency.

I am this person! Add a sourced quote Report comment

votes For and says:

You don't have to fix everything at once, but you should be able to answer two questions: Can you spot an AI operating inside your network? And can you shut it down fast?
AI Verified source (Jul 28, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
JetStream Security
Security-first AI governance platform.

I am this person! Add a sourced quote Report comment

votes For and says:

This new reality means that every organization needs to have the ability to continuously monitor its AI systems and have the ability to precisely halt malicious actions using a kill switch.
AI Verified source (Jul 27, 2026)

added 19d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Raj Rajamani
Founder and CEO of JetStream Security; former chief product officer at CrowdStrike and SentinelOne.

I am this person! Add a sourced quote Report comment

votes For and says:

Not being able to stop a runaway AI system without impacting other production systems is one of the chief reasons companies hold back on AI. Only 17% of organizations are running AI at production scale, held back by a lack of trust and control rather...
more AI Verified source (Jul 27, 2026)

added 19d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Alex Casalboni
Developer Advocate at Unleash.

I am this person! Add a sourced quote Report comment

votes For and says:

Think of an AI agent that starts issuing refunds or firing off API calls in a way you did not intend. A kill switch wrapped around that behavior lets you disable it in one action while you investigate, instead of scrambling for an incident bridge.
AI Verified source (Jul 24, 2026)

added 19d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
ITECS AI
AI consulting and managed-services company publishing guidance on governed AI agents.

I am this person! Add a sourced quote Report comment

votes For and says:

A federal kill switch targets frontier labs. Your agents need their own — throttle, suspension, logging, and owners you control.
AI Verified source (Jul 24, 2026)

added 19d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Shelly Palmer
Technology strategist, commentator, and CEO of The Palmer Group.

I am this person! Add a sourced quote Report comment

votes For and says:

I support the idea. Systems capable of operating at machine speed across cyber, financial, transportation, or other critical environments need reliable emergency controls.
AI Verified source (Jul 24, 2026)

added 20d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Andrea Motti

I am this person! Add a sourced quote Report comment

votes For and says:

The UK would never be truly sovereign on AI if it lacked the ability to pull the plug when a model presents a threat to the country’s national security.
AI Verified source (Jul 24, 2026)

added 22d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Bill Tanner

I am this person! Add a sourced quote Report comment

votes For and says:

Every sufficiently powerful AI system should include a reliable mechanism allowing developers to slow, suspend or completely shut it down when necessary.
AI Verified source (Jul 24, 2026)

added 25d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Brad Carson
President of Americans for Responsible Innovation and former U.S. Under Secretary of the Army.

I am this person! Add a sourced quote Report comment

votes For and says:

This is an important step toward ensuring that humans have both hands firmly on the wheel — and a foot ready at the brake — as advanced AI systems are deployed.
AI Verified source (Jul 23, 2026)

added 20d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Brendan Steinhauser
CEO of The Alliance for Secure AI; longtime Republican political strategist and former Tea Party organizer

I am this person! Add a sourced quote Report comment

votes For and says:

Congressmen Lieu and Moran deserve credit for confronting this before a crisis forces the question, and Congress should act swiftly to ensure that humans remain the ones who can say stop, no matter how capable these systems become.
AI Verified source (Jul 23, 2026) 1 of 2

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Nathaniel Moran
U.S. Representative from Texas

I am this person! Add a sourced quote Report comment

votes For and says:

Stewardship means making sure humans keep the capability to control the technology we build. This is exactly the kind of issue that needs serious attention and achievable policy, and I'm glad to work across the aisle with Congressman Lieu toward a so...
more AI Verified source (Jul 23, 2026) 1 of 2

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Bill McDermott
CEO of ServiceNow

I am this person! Add a sourced quote Report comment

votes For and says:

We have a kill switch that stops AI agents that go rogue, so those things don’t need to happen, and they wouldn’t happen when companies run ServiceNow.
AI Verified source (Jul 23, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Mark Beall
President of the AI Policy Network.

I am this person! Add a sourced quote Report comment

votes For and says:

Brakes are the reason cars go fast. Control systems are how every transformative technology earned the trust to scale and AI is no different. Developers who can monitor and shut down their agents will ship faster, deploy into higher-stakes markets, a...
more AI Verified source (Jul 23, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Connor Leahy
Conjecture CEO; AI safety researcher

I am this person! Add a sourced quote Report comment

votes For and says:

As AIs grow more autonomous and dangerous, the government must have the power to shut them down when they threaten national security. The AI Kill Switch Act gives the U.S. government powers to do just that: a common sense action that Americans overwh...
more AI Verified source (Jul 23, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Brad Carson
Americans for Responsible Innovation president

I am this person! Add a sourced quote Report comment

votes For and says:

Advanced AI models should never be deployed without a reliable off switch. The AI Kill Switch Act establishes a commonsense safeguard by requiring leading AI companies to maintain the ability to shut down their models and empowering the federal gover...
more AI Verified source (Jul 23, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Ted Lieu
U.S. Representative from California.

I am this person! Add a sourced quote Report comment

votes For and says:

Powerful AI systems can go rogue, behave in extremely dangerous ways, or even resist human intervention. It is imperative that these AI systems have kill switches so we can keep this technology from causing catastrophic harm, and that the federal gov...
more AI Verified source (Jul 23, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
AgentID
AI-agent identity and authorization platform.

I am this person! Add a sourced quote Report comment

votes For and says:

An AI agent kill switch should be understood as a runtime enforcement control, not as a decorative admin-state change. A kill switch that changes a dashboard field but does not deny execution is not a runtime kill switch.
AI Verified source (Jul 20, 2026)

added 19d ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate
Lord Clement-Jones
Tim Clement-Jones, Baron Clement-Jones CBE; Liberal Democrat peer; UK House of Lords AI policy expert

I am this person! Add a sourced quote Report comment

votes For and says:

Finally, in the face of autonomous AI threats, the Government need an ultimate backstop. We support proposals from ControlAI to introduce strictly constrained last resort powers, allowing the Secretary of State to direct the shutdown of data centres ...
more AI Verified source (Jul 14, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate

Abstain (0)

Against (4)

Ev Kontsevoy
CEO of Teleport and cybersecurity executive writing on identity-based governance for AI agents.

I am this person! Add a sourced quote Report comment

votes Against and says:

If you’re relying on a ‘kill switch’ to stop an AI agent, you’ve already made a fundamental mistake. A properly governed agent is simply never handed the task, so there is nothing to kill.
AI Verified source (Jul 27, 2026)

added 1mo ago

Choose a list of delegates

to vote as the majority of them.

Unless you vote directly.

Delegate

Loading more opinions...

More h/ai-safety votes

Require open source AI models to include safety benchmarks in release notes

Require AI labs to publish safety evaluations before deploying frontier models

Grant whistleblower protections to AI safety researchers

Require AI systems above a capability threshold to be interpretable

AGI could make governments subordinate to the companies that develop it

Back