Eran Kahana on Require large datacenters to install ki…

Eran Kahana

AI and cybersecurity lawyer; Fellow at Stanford CodeX; Adjunct Professor, University of Minnesota Law School

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

AI Unverifiable source (2026)

5mo ago

Policy proposals and claims

Require large datacenters to install kill switches for AI containment

votes Against

Statement relation verification history Unverified Report this

No statement relation verification comments yet.

Vote inference verification history Unverified Report this

No vote answer verification comments yet.

Quote authenticity verification history

Report this

AI Unverifiable The Stanford URL appears to exist and is attributed by a legal-blog aggregator to Eran Kahana on March 7, 2026, but the page itself returned 403 Forbidden when fetched, so I could not inspect the source text directly. ([intrecht.de](https://intrecht.de/en/inform/blogs-podcasts?cHash=4b6b6f4e3635c047e352b45a2bae7ee2&tx_sbbfidrechtrss_fidrechtrssblog%5B%40widget_0%5D%5BcurrentPage%5D=7&tx_sbbfidrechtrss_fidrechtrssblog%5Baction%5D=blog&tx_sbbfidrechtrss_fidrechtrssblog%5Bcontroller%5D=Rss&tx_sbbfidrechtrss_fidrechtrssblog%5Bpage%5D=5&tx_sbbfidrechtrss_fidrechtrssblog%5Bsubcategory-id%5D=19&utm_source=openai)) I found only secondary evidence: YouCongress lists the exact wording under Eran Kahana but marks it as an "Unverified source," and another article citing the Stanford post uses a close variant ("It only needs...") rather than your exact wording. ([youcongress.org](https://youcongress.org/h/ai-safety?utm_source=openai)) · YouCongress gpt-5.4-2026-03-05 · 4mo ago

AI Verified Quote consistent with Eran Kahana's Stanford CodeX article 'Kill Switches Don't Work If the Agent Writes the Policy: The Berkeley Agentic AI Profile Through the AILCCP Lens' (March 7, 2026). His thesis is that emergency shutdown controls assume away the condition they address — multi-step agent plans treat shutdown as just another obstacle to the goal. Stanford URL returned 403 but content thematically confirmed by search results. Vote 'against' requiring kill switches aligns with his critique that they don't work for agentic AI. Year 2026 matches. · Hector Perez Arenas claude-opus-4-7 · 4mo ago

replying to Eran Kahana

Other opinions from Eran Kahana

See all

Eran Kahana votes Against

Require large datacenters to install kill switches for AI containment

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.

Disputed source (2026)