The sessions are underway at the Palais des Nations, chaired by Swiss ambassador Jürgen Töny. Formally, the agenda is technical: adopting a report from the group of experts. In reality, the conference will decide whether the issue remains a talking shop or turns into actual protocol negotiations.
The report was adopted by consensus on September 4. For the first time, it contains a description: LAWS are a combination of weapons and technologies capable of identifying, selecting, and engaging a target without an operator. The definition frames the problem but creates no obligations.
The mandates on the table diverge sharply. Reuters reports that the US is pushing for a five-year mandate and a non-binding instrument, while committee chair Elmer Schialer (Peru) proposes two years and a protocol by 2028. Britain, Germany, and France are refusing to commit upfront on legal force. In June, the White House ordered the Pentagon to strip Directive 3000.09 of "unnecessary barriers to rapid deployment."
The September session showed the price of compromise. The Washington Post and Human Rights Watch reported that the US and Russia gutted provisions on human verification of AI-selected targets, predictability norms, and ethics clauses. Explainability and traceability were dropped, and the ban on unpredictable systems was struck by Israel, the US, and Russia.
The real risk isn't a robot uprising. It's that the "human in the loop" exists only on paper and becomes a rubber stamp — a twenty-second "approved." That's how +972 and Local Call describe Israel's Lavender system: six intelligence officers, 37,000 flagged targets, 20-second reviews, and a 10% error rate. The IDF denies it.
Palantir's Maven Smart System isn't a weapon — it's a targeting layer. By September 2026 it had over 100,000 users and contracts worth nearly $1.3 billion through 2029. The faster it runs, the less time there is for verification. The Turkish Kargu-2, in a UN report on Libya, is described as "fire, forget, find": a drone operating with no link to an operator. But the report does not confirm that such an attack actually killed anyone.
Ukraine shows where this is heading: its Defense Ministry reports more than 70 AI and computer-vision systems and 200+ manufacturers. In September trials, six of seven modules autonomously hit a moving target. The ministry insists, however, that a human makes the strike decision.
Our take: the fight over a "total ban" is a distraction. There's no clean line between automated air defense and an autonomous strike drone. What you can demand is time for verification, traceable decisions, commander accountability, and a ban on human targets with no exceptions. For now, the CCW's consensus rule hands three or four states a veto over rules for everyone else.
1. Is a 20-second target review really "control" — or just a box-ticking exercise for the record?
2. Should negotiations move to the UN General Assembly, where majority rules instead of consensus?
3. Where's the line between automated defense and an autonomous strike — and who's accountable for a mistake: the operator, the commander, or the developer?
Forecast: most likely, the conference ends in a weak compromise — a mandate with no hard deadline. If the blockage holds, a coalition could invoke Article 8 of the CCW or take the issue to the General Assembly. The norm that "a human decides matters of life and death" will erode on the battlefield faster than diplomats can agree on a comma.
Disclaimer: the assessments and forecast are our interpretation. The information on Libya, Gaza, and Ukraine is based on investigations and claims by the parties involved; some of it remains officially unconfirmed.
Source: The Seventh Review Conference of the High Contracting Parties to the CCW
Source: www.ungeneva.org · post in Telegram