From the CrowdStrike edition of November 7, 2026
AI Risk Today: CrowdStrike's kill switch only cuts gateway-routed agents for now; Forrester and IDC urge testing, and federal kill-switch bills target AI developers, not users
AI Risk Today · Independent AI policy and risk reporting · United States
- Publisher
- AI Risk Today
- Original publication
- November 4, 2026
- Source retrieved
- 2026-11-08
- Event date
- 2026-11-03
What happened
AI Risk Today reports that the kill switch currently works by cutting an agent's connection to its model through Falcon Guardian and cannot stop agents that connect another way; endpoint stopping comes later with no date, and Security Boulevard reported blocking restarts on computers is also a later version. Forrester's Jeff Pollard told Forbes stopping an agent does not reverse earlier actions and agents may route around a switch; IDC's Michele Abraham said companies should test a kill switch before they need it, including cutoff of delegated agents and restart. The article notes the House AI Kill Switch Act and Senate AI Emergency Button Act would apply only to AI developers, and that OpenAI recommended in October that businesses using agents be able to stop them.
Why it matters
Gives state and local policymakers and CIOs a plain-language view of current kill-switch scope and the federal legislative backdrop, useful when drafting agency AI policies that require a stop capability for deployed agents.
Evidence and measured results
Opened article summarizing CrowdStrike claims, secondhand analyst comments via Forbes and Security Boulevard, and status of two federal bills.
Limitations and uncertainty
Secondhand analyst quotes; page partially blocked to automated curl (opened via fetch tool); legislative status as of publication; no independent test of the product.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-11-08; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Agency AI policies demand a way to stop deployed agents, but current tools cover only some agents.
- Stakeholders
- State CIO/CISO
- Agency AI governance board
- Legislative liaison
- Counsel
- Discovery questions
- Does your AI policy require a stop capability?
- Which agents connect directly to models?
- How will you test stop and restart?
- Value hypothesis
- Tested, documented stop controls can satisfy agency policy requirements for sanctioned agents.
- Potential engagement
- Policy-to-control mapping and kill-switch test plan for sanctioned agents.
- Unsupported claims
- That federal law requires agencies to have kill switches
- That all agents can be stopped today
Pre-sales engineering
Role takeaway
- Fit
- Partial: covers gateway-routed agents today.
- Architecture and integration
- Route sanctioned agents through the AI Gateway; flag direct connections as policy exceptions.
- Prerequisites
- Gateway routing
- Agent registry
- Test tenant
- Constraints
- No date for endpoint stopping
- Agents may route around controls
- Security
- Verify delegated-agent cutoff and controlled restart.
- Proof of value
- Parent-agent stop cuts delegated agents
- Restart requires authorization
- Direct-connect agents detected as exceptions
Delivery
Role takeaway
- Work
- Coverage register and acceptance tests aligned to agency AI policy.
- Dependencies
- Agency AI policy
- Guardian entitlement
- Endpoint roadmap
- Ownership
- Agency AI governance lead with CISO.
- Skills and adoption
- Policy staff and engineers jointly review stop/restart tests.
- Governance checkpoints
- Policy requirement defined
- Tests passed
- Exceptions approved
- Acceptance criteria
- Proposed: every sanctioned agent listed as stoppable or as an approved exception
- Proposed: restart test passes
- Risks
- Policy overstates coverage
- Direct-connect agents ignored
- Legislative assumptions outdated
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Agents that bypass the Guardian gateway are out of scope until endpoint enforcement ships.
Governance
Who approves, reviews and stays accountable for outcomes?
Federal bills would not cover state or local agencies as deployers; agency policies must set their own stop and restart requirements.
Security and privacy
What data, permissions and controls need testing?
Testing must confirm delegated subagents are cut off and that restart is controlled.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Plain-language explanation is useful for non-technical policymakers.
Procurement
What should contracts, pricing and exit terms secure?
Write stop and restart tests into acceptance criteria for AI security purchases.
Operating model
Which teams own the service once it runs?
Agencies, not federal law, will set deployer stop requirements for now.
Publication history
- 2026-11-07CrowdStrike · Issue 105 resources
Stable resource ID: airisktoday-crowdstrike-kill-switch-two-limits-2026-11-04