US-China experts warn autonomous AI could force minute-long nuclear decisions; propose red lines and hotline

A multi-national group of US and Chinese security experts warned that increasingly autonomous military AI could compress the time available for nuclear decisions to minutes, and urged nuclear-style safeguards to prevent inadvertent escalation.

The experts recommended clear 'red lines' limiting autonomous actions in critical systems, verification standards to ensure compliance, and establishing a US-China hotline to manage crises involving autonomous weapons (per news.google.com).

Advocates of immediate technical measures argued that enforceable engineering constraints and testing regimes can reduce the risk of accidental launches or rapid escalation; skeptics within the group said such measures need political-level agreement first, because technical fixes alone cannot bind militaries or guarantee trust (per news.google.com).

The report frames the danger concretely: faster decision loops created by AI can outpace human deliberation and crisis diplomacy, which is why participants pressed for rules modeled on nuclear arms control even as they acknowledged verification challenges (per news.google.com).

Proponents pointed to practical steps — interoperable verification tools, audit logs for autonomous systems, and narrow prohibitions on fully autonomous targeting in strategic domains — as the most immediately actionable items; critics warned that without reciprocal political commitments between Washington and Beijing, those tools could be ignored or gamed (per news.google.com).

The experts also elevated communications measures, recommending a dedicated hotline between the United States and China specifically for incidents involving autonomous military systems, intended to mirror crisis hotlines used for nuclear risks and reduce chances of miscalculation (per news.google.com).

If implemented, the package would shift how both governments govern emerging military AI: from largely technical development toward binding limits, monitoring, and crisis-management channels — but the report makes clear that achieving mutual confidence and verification remains the central hurdle (per news.google.com).

• The single source presents both the technical-implementation framing and the political-consent framing; it notes proponents urging immediate technical rules and skeptics arguing political agreements must come first (per news.google.com).

• The source does not quantify how often expert participants favored technical steps over political agreements, leaving the balance between the two approaches unclear (per news.google.com).

• No source text in this pack mentions prior concrete incidents where autonomous systems directly affected nuclear-launch decisions;
• No source text mentions civilian casualty estimates tied to autonomous weapons incidents;
• No source text cites existing legal or treaty frameworks (e.g., specific arms control treaties) that would govern these proposed safeguards;
• No source text names specific defense contractors, procurement programs, or budgets that would be affected by red lines.

• The source does not provide differing numerical estimates for decision-time compression; no alternative figures are given (per news.google.com).

• All descriptions link the risk to increasing autonomy in military AI; the source does not dispute who acted first but frames the issue as a developing technological trend rather than a single triggering event (per news.google.com).

• The proposals and warnings are attributed to a group of US and Chinese security experts; the source does not name individual authors or institutions (per news.google.com).