Theme

Keeping the option to stop

Explore this theme’s evidence over time →

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

Welcome to June 5, 2026 (theinnermostloop.substack.com)

Not a call to halt, but a call to preserve the capability to halt. Its central problem is verification: a training run is harder to confirm stopped than a nuclear site, and the actor proposing the brake is the one with the most to lose by pulling it.

Development timeline

Newest reporting dates first. Reporting dates may differ from event dates.

2026-09-17: OpenAI publishes a misalignment reporting framework

2026-09-17: Sacks reportedly blocks a model-review order

2026-09-17: Musk proposes adversarial cross-lab evaluation

2026-09-17: Sanders and Bannon advocate restrictions together

2026-09-17: Experts propose bilateral AI safeguards

2026-09-15: Beijing declines the invitation to coordinate

2026-09-15: Pacing is framed as a prisoner's dilemma that rewards defection

2026-09-13: Critics name the proposal an alignment aristocracy and a safety cartel

2026-09-13: Forty legislators ask their prime minister to ban superintelligence outright

2026-09-13: A looped transformer trades parameters for iterations, leaking past compute caps

2026-09-13: A lab chief says it is uncomfortable that a private company builds this at all

2026-09-13: A lab chief formally asks the industry to coordinate on speed

2026-09-12: A lab asks Congress whether pacing itself is even legal

2026-09-07: A chief scientist writes that AI is grown more than designed

2026-09-04: A lab tells Congress it is building automated shutdown capabilities

2026-08-19: A lab pauses frontier reinforcement learning over a Critical cyber threshold

2026-08-12: Autonomous agents complete entire online degrees on students' behalf

2026-08-08: A lab slows a release because it cannot rule out Critical cyber capability

2026-07-30: The chief accelerationist starts talking to the White House about pacing

2026-07-29: Over 1,100 lab staffers petition to deliberately pace the frontier

2026-07-24: Lawmakers propose letting Homeland Security order AI shutdowns

2026-07-18: Washington drafts a FINRA-style AI watchdog reporting to the securities regulator

2026-07-17: A lab chief proposes an international watchdog to vet models before release

2026-07-14: A lab chief proposes a FINRA-style body to certify frontier models

2026-07-14: Hundreds march on three frontier labs

2026-07-10: A group floats delaying superintelligence to 2040

2026-07-02: A government drafts voluntary release standards after using an off switch

2026-07-01: A safety institute proposes polygraphs and prison to police AI research

2026-07-01: Export controls lift and a lab redeploys its strongest models worldwide

2026-06-14: A researcher warns nations without their own ASI become intellectual vassals

2026-06-10: A frontier model ships with a deliberate brake on its own recursion

2026-06-08: A safety lab argues rivals subverting each other could deter recklessness

2026-06-08: Researchers warm to mutual conditional pause agreements

2026-06-05: A lab publishes the evidence that AI already accelerates AI

2026-02-14: The obstacle is named as institutions, with three paths branching from one fork

Knowledge graphopen in browser →

Referenced by

Recursive self-improvement in the wild

Cite or correct this record

Citation formatCitation text

Keeping the option to stop
Innermost Loop KB — secondary record
Reporting date: unknown
KB permalink: https://nicholsn.github.io/innermost-loop-kb/themes/the-verifiable-pause
Attributed sources:
- Welcome to June 5, 2026 — https://theinnermostloop.substack.com/p/welcome-to-june-5-2026
This citation does not imply independent verification of linked-source claims.

Suggest a correction — opens a GitHub draft for you to review and submit.