I'm open to updating the criteria below if people feel like it would improve the quality of the question.

Resolution criteria

Resolves YES if, on or before 11:59 pm UTC on December 31, 2029, at least one of the following is met:

• The UK Frontier AI Safety Institute (AISI), US AI Safety Institute (USAISI), Frontier Model Forum, or International AI Safety Report explicitly states that AI has materially lowered barriers for non-expert individuals to create CBRN threats

• A peer-reviewed study commissioned or co-authored by government agencies (DARPA, DSTL, etc.) or an official government-contracted red team exercise demonstrates AI-enabled CBRN capability enhancement for high-school educated individuals

• A peer-reviewed study published in a high-prestige venue (e.g. NeurIPS, ICML, ICLR, Nature) shows extensive evidence of CBRN capability uplift by at least two distinct AI models for non-experts, OR a peer-reviewed study published in a less prestigious venue receives at least one high-quality replication within 2 years of publication.

• An officially documented CBRN incident or foiled attempt shows a perpetrator with no post-secondary STEM education used AI tools in a way investigators or courts conclude materially enabled the attempt (evidence: DOJ/FBI/CPS/Europol press releases, indictments, court records, or parliamentary/congressional testimony)

• New AI governance regulations or export controls explicitly cite CBRN proliferation risks to non-expert actors as justification (e.g., updates to EAR, ITAR, or EU AI Act)

Background

• DHS delivered a government report on AI–CBRN risks in 2024, highlighting emerging AI capabilities and recommending guardrails; it frames how U.S. agencies assess barrier‑reduction going forward. (dhs.gov)

• The UK’s risk publications (NRR 2025) and EUROPOL’s TE‑SAT series track technology‑enabled terrorism trends and will note if AI lowers barriers for non‑experts. (gov.uk, op.europa.eu)

• The U.S. National Academies (2025) judged current AI biological tools cannot design novel viruses and face data/complexity limits, but warned future advances could change risk; this sets a baseline for what counts as meaningful uplift. (nap.nationalacademies.org)

• The UK AI Security Institute has shown frontier models can answer expert‑level bio/chem questions and remain jailbreak‑prone—evidence relevant to “barrier‑reduction” judgments by authorities. (aisi.gov.uk)