Governance and control

Hendrycks proposes deterrence and nonproliferation for superintelligence

Hendrycks, Eric Schmidt, and Alexandr Wang propose a strategy combining deterrence against destabilizing AI projects, controls on chips and model weights, technical misuse safeguards, legal rules for agents, and measures to limit automation-driven instability.

0 comments · 0 votesOpen discussion

Public discussion is readable by everyone. Sign in to comment, reply, or vote.

No comments yet. Start the discussion.

CURRENT ASSESSMENT · REVISION 1
AWAY FROM DOOM22confidence 58/100

Why it moved the index

The proposal adds specific international-control mechanisms for advanced AI, including compute tracking, model-weight security and deterrence, but their effectiveness and escalation risks remain untested policy judgments.

AUDIT TRAIL

Assessment history

  1. R1
    Away 22 · confidence 58

    Initial inclusion from a fully read first-person policy synthesis with mechanisms distinct from Hendrycks's earlier risk taxonomy.

    13 Aug 2026
SHARE THE FINDINGS

Share this page

DoomBench social sharing card for Hendrycks proposes deterrence and nonproliferation for superintelligence.
  1. DoomBench assesses “Hendrycks proposes deterrence and nonproliferation for superintelligence” as evidence moving away from doom, with magnitude 22 and confidence 58 out of 100 in the governance and control category.

  2. The DoomBench assessment of “Hendrycks proposes deterrence and nonproliferation for superintelligence” is based on reporting from Center for AI Safety and records the editorial rationale, source quality, attribution, and revision history.

  3. DoomBench summarizes “Hendrycks proposes deterrence and nonproliferation for superintelligence” as follows: Hendrycks, Eric Schmidt, and Alexandr Wang propose a strategy combining deterrence against destabilizing AI projects, controls...

    https://www.doombench.com/news/hendrycks-proposes-deterrence-and-nonproliferation-for-superintelligence-2025-03-06