A forum, launched with a warning

The launch itself is procedural: a 105-minute side event on Thursday 22 October 2026, 13:00–14:45 EDT in Conference Room 4 at UN Headquarters, run with Switzerland and Pakistan as a side event to the 81st UN General Assembly First Committee. The scheduled programme is welcome remarks, a moderated exchange, and an invitation for organisations to register interest in joining. The scheduled speakers are Oliver Höhne, Switzerland's deputy permanent representative to the UN; Muhammad Usman Iqbal Jadoon, Pakistan's deputy permanent representative; Izumi Nakamitsu, the UN Under-Secretary-General and High Representative for Disarmament Affairs; and UNIDIR Director Robin Geiss, with UNIDIR researcher Yasmin Afina moderating. The Centre behind it has existed since June 2026. Nothing about Thursday creates an obligation for any state or company. The launch opens expressions of interest for organisations to participate; it is not itself a membership agreement or a new treaty process.

The briefing that preceded it is the substantive part. Geiss told reporters that AI and autonomy are already embedded widely across the military domain, in intelligence analysis and target identification, drones, cyber attacks and information operations. Military competition is pushing adoption in a specific direction: less time between identifying and engaging a target, and more targets engaged per day. Speed is the point of the procurement, not a side effect.

Geiss warned that AI is compressing military decision cycles and narrowing the time available for human judgment, including in situations involving nuclear weapons. In his words, "the margin for human judgment is increasingly narrow, including in the nuclear domain." He added that opaque systems raise the risk of error, miscalculation and unintended escalation, and that falling entry costs mean sophisticated AI-supported operations are coming within reach of non-state actors as well. He also noted that a small number of countries and companies are shaping security outcomes "in ways that traditional arms control frameworks were not designed to address".

The argument is about the process, not the trigger

Most public argument about military AI stops at the weapon: whether a machine may select and engage a target on its own. Geiss drew a wider line. A system does not need to fire to shape a decision. The tools that gather intelligence, rank targets, filter alerts and summarise a situation decide what a commander sees and how much time appears to be available.

That is the risk he is pointing at. A human may retain formal authority over the final decision while relying on AI-generated intelligence, rankings and alerts whose assumptions are difficult to examine under time pressure. In that situation, human approval alone may not guarantee meaningful control over the decision-making process. It is a risk argument about the process that produces a recommendation, not a documented finding that control has already been lost in any particular system — and it reaches well beyond whether an autonomous weapon can select and engage a target.

That distinction is why the Centre's remit is broader than the track most governments are actually negotiating on. Under the Convention on Certain Conventional Weapons, government experts agreed in September 2026 on elements for a possible future instrument on lethal autonomous weapons, and the convention's Seventh Review Conference in November 2026 will consider the next steps, including possible negotiations on a binding text. Even a concluded instrument of that kind would govern autonomous weapons. It would not, by itself, govern the decision-support systems Geiss is describing.

What this week actually produces

Three of the Centre's initiatives move in the same week as the Forum launch, and all three are voluntary. Targeted consultations are opening on a framework for responsible industry behaviour, which UNIDIR says will provide practical, voluntary guidelines grounded in international law and the UN Guiding Principles on Business and Human Rights, and is meant in time to inform government procurement and public-private partnerships; Geiss told the briefing that UNIDIR is partnering with the Office of the High Commissioner for Human Rights on the framework. A consultation opening is not a framework adopted: no company is bound by anything this week. A lexicon project launches on Friday 23 October, because states currently use words such as autonomy, safety and security to mean different things, and the ambiguity ends up inside national positions. A third initiative, with UNITAR, examines misuse of AI by non-state armed groups and looks for gaps in existing non-proliferation and export-control frameworks.

Geiss also pointed to one concrete step between rivals. According to his 20 October briefing, the United States proposed an AI safety hotline with China last month, for an AI incident that could threaten national security. The briefing record reviewed for this article does not establish whether China accepted the proposal, or whether such a hotline is operating. He said it in the same briefing in which he described the Secretary-General's repeated calls for stronger regulation of military AI and a legally binding treaty on lethal autonomous weapons as finding a growing echo, and geopolitical competition as a serious obstacle to the way forward. Both halves are in the record.

Sources

  1. [1] UN Transcripts (transcripts.un.org), automatic speech-recognition transcript of the press conference "Robin Geiss (UNIDIR) on the launch of a range of new initiatives from UNIDIR's Centre of Excellence on AI, Peace and Security", 20 October 2026 — read directly for this article: the two direct quotations from Geiss were verified word-for-word against this transcript; it is an automatic transcript, not an official UN recordRead source
  2. [2] UN UNifeed, "UN / AI Peace and Security Presser", record of the 20 October 2026 briefing (the Forum launch and the "only multi-stakeholder body" description; Geiss on embedded AI, compressing decision cycles and the narrowing margin for human judgment including in the nuclear domain; opacity, error and escalation risks; the non-state actor threshold; the arms-control frameworks passage; the US–China hotline proposal; the industry framework, lexicon and non-proliferation initiatives; the Centre's June 2026 launch)Read source
  3. [3] UNIDIR, "UNIDIR's Centre of Excellence on AI, Peace and Security: Forum launch", official event page (Thursday 22 October 2026, 13:00–14:45 EDT, Conference Room 4, UN Headquarters; side event to the 81st UN General Assembly First Committee, in partnership with Switzerland and Pakistan; scheduled speakers Oliver Höhne, Muhammad Usman Iqbal Jadoon, Izumi Nakamitsu and Robin Geiss, moderated by Yasmin Afina; expressions of interest; the Centre's June 2026 launch and purpose)Read source
  4. [4] Dig.watch (Digital Watch Observatory), "UNIDIR warns of declining human control in military AI", published 21 October 2026, updated 21 October 2026 (identify-to-engage speed and targets-per-day detail; the distinction between autonomous weapons and wider decision-making; industry framework partnership with the Office of the UN High Commissioner for Human Rights; lexicon and UNITAR non-state misuse initiative; CCW September 2026 agreed elements and the November 2026 Seventh Review Conference context)Read source
  5. [5] The Tower Post, "UN Military-AI Initiative Puts Human Control at the Centre of a New Global Forum", 21 October 2026 (Associated Press-based cross-check of the briefing and the Thursday launch)Read source