Pause commitment

In plain English

Pauses come in two forms. Lab commitments are written into safety frameworks: if a model reaches a capability level the lab cannot yet secure, development or deployment stops until mitigations are ready. Public pauses are calls for a whole industry to stop, such as the 2023 open letter asking for a six month pause on training systems more powerful than GPT-4, or the 2025 statement calling for a prohibition on building super intelligence until it can be done safely and with strong public buy-in.

Lab pause commitments have weakened in several cases. Anthropic replaced an earlier pause pledge with commitments that depend on what competitors do, and Meta changed its response at the Critical threshold from stop to develop with mitigations. Microsoft's framework commits it to pause development and deployment of a model whose risk it cannot sufficiently mitigate. The Containment Board's shutdown and pause grade measures how firm each promise is.

Why it matters for rogue SI

A pause is the planned version of a kill switch: stopping before a system is built rather than after it misbehaves. Its credibility decides whether labs would stop in time.

Laws and rules that use it

• Stalled or withdrawn SB 24-205 Colorado
• Voluntary or guidance FLI Statement Global civil society

Sources

1. The pause letter (2023) Future of Life Institute
2. The Statement on Superintelligence (2025) Statement on Superintelligence
3. Colorado Delays AI Act to June 2026; Businesses Get Breathing Room PYMNTS · Sep 10, 2025
4. Statement on Superintelligence Future of Life Institute