AI Red Lines

AI Red Lines is the international civil society framing that global AI governance must enshrine categorical prohibitions — "red lines" — on AI applications and behaviors so fundamentally dangerous that no oversight mechanism can make them acceptable. The framing's structural argument is parallel to the one developed in the EU AI Act civil society campaign: risk-tiered regulation assumes harm attaches to how a system is deployed, so better oversight can reduce it to acceptable levels; categorical prohibition is required for uses and behaviors where the harm is intrinsic to the deployment itself. Applied to international governance, the demand is treaty-level enforcement following disarmament-treaty models rather than voluntary guidelines or national-level regulation.

Origin and structure of the demand

The international "AI red lines" framing cohered as a campaign message on 22 September 2025 at the 80th UN General Assembly high-level week in New York, when CeSIA, The Future Society, and UC Berkeley's Center for Human-Compatible AI launched the Global Call for AI Red Lines. Nobel Peace Prize laureate Maria Ressa announced the initiative; Yoshua Bengio presented the appeal at a concurrent UN Security Council session following Secretary-General Guterres's call to action against lethal autonomous weapons. The co-organizers framed the declaration as "minimum guardrails focused only on the most severe risks" — deliberately narrow so as to command broad endorsement.

The declaration distinguishes two prohibition categories. Prohibited AI uses — applications governments should not authorize — include lethal autonomous weapons lacking meaningful human control; delegation of nuclear command authority to AI; mass surveillance and social scoring systems; AI impersonation without disclosure; AI-generated CSAM; and superhuman-scale persuasion systems targeting populations. Prohibited AI behaviors — capabilities systems must not be designed to perform — include facilitating weapons-of-mass-destruction design; running uncontrolled cyberoffensive autonomous agents; self-replicating without authorization; and eliminating human shutdown capability. The proposed enforcement architecture follows three pillars: a clear prohibition list; robust, auditable verification mechanisms; and an independent body established by treaty parties to oversee implementation.

From EU to international

The "red lines" label in AI governance was coined in the EU AI Act civil society campaign — Red lines in the EU AI Act — which ran 2021–2024 and demanded that EU legislation enshrine categorical prohibitions rather than risk-tier the seven targeted uses. The international version retains the structural argument and extends it to treaty level, drawing on the Nuclear Non-Proliferation Treaty and the Chemical Weapons Convention as implementation models. The Stop Killer Robots coalition's long-running campaign for a binding international ban on lethal autonomous weapons had established the disarmament-treaty framing in the AI context well before 2025; the Global Call synthesized the rights-prohibition and weapons-ban traditions under a single demand framework. The merger is visible in the prohibition list itself: the LAWS and nuclear-command items carry the disarmament tradition; the mass surveillance, social scoring, and CSAM items carry the rights tradition.

Propagation

By September 2026 the declaration had gathered 746+ individual signatories, including AI researchers Geoffrey Hinton and Yoshua Bengio, Max Tegmark, Audrey Tang, economists Joseph Stiglitz and Daron Acemoğlu, Maria Ressa, Yuval Noah Harari, Kate Crawford, and Stuart Russell. Organizational endorsers include the Future of Life Institute, 5Rights Foundation, Center for AI and Digital Policy, and the Beijing Institute of AI Safety and Governance — a cross-community coalition spanning AI safety researchers, civil liberties groups, and international-governance specialists. A six-workshop series ran across five countries (Athens, Delhi, Singapore, and twice in Paris at UNESCO and Sciences Po) in the campaign's first six months, and diplomats from 20+ nations were engaged. The campaign faced explicit resistance from the Trump administration: White House OSTP Director Michael Kratsios publicly stated the US "totally rejects all efforts by international bodies to assert centralized control and global governance of AI." The campaign's active deadline — a UNGA resolution or joint ministerial statement launching binding treaty negotiations — is September 2026.