Military AI: UN consultation opens before the rules are written

Military AI is entering a new phase of UN discussion: the choices made upstream by industry. On 23 October in New York, the United Nations Institute for Disarmament Research (UNIDIR) is scheduled to present questionnaires to help develop a conduct framework for companies. The proposal is still under consultation and is intended to be voluntary. [3][4]

A consultation, not an adopted rule

The meeting scheduled for 23 October is part of the General Assembly First Committee’s work. UNIDIR’s public programme says three questionnaires will target industry, States and civil society. They are intended to map current practices and gather expectations on responsibility and compliance. [3]

The process builds on a multi-year programme launched in February 2026 with the Office of the UN High Commissioner for Human Rights. Its aim is to establish common expectations for companies and help governments clarify responsible procurement and public-private partnerships. UNIDIR describes the framework in development as practical and voluntary. [3][4]

Why industry is central

Choices that shape a system are made long before an armed force uses it. Training data, test scenarios, autonomy thresholds, technical logs and maintenance are often determined by companies that design, integrate or supply the tools. UNIDIR says these choices can affect reliability, predictability and the ability to establish what happened after an incident. [4]

The issue extends beyond software designed exclusively for the military. AI crosses sectors: related tools can support logistics, image analysis, cybersecurity or the detection of deceptive content. That civil-defence intersection merits attention; by itself, it does not classify a product as a regulated dual-use item. [4]

The Associated Press reported on 20 October that the initiative is intended to bring together participants from multiple sectors. It also noted questions about AI in early warning, intelligence and decision support, including human control and accountability. UNIDIR’s proposed Forum is meant to provide a space for exchange and consultation, open to organisations that express interest. [1][2][5]

What the initiative does not settle

A forum does not have regulatory powers. The 22 October programme describes a place for dialogue among governments, international organisations, companies, researchers and civil society; it creates neither new obligations nor a certification process. Its influence will depend on who takes part, how transparent the consultations are and whether States use the proposals. [1][2]

The industry framework itself is still being written. Questionnaires are meant to identify current practices and disagreements first. UNIDIR says States retain their legal responsibilities when acquiring and using these systems. Voluntary commitments could help document testing and allocate roles, but they do not replace applicable law or public accountability. [3][4]

The test will be what follows

The practical questions concern the system lifecycle: what test evidence should suppliers be able to share? How should failures be reported, records preserved and responsibilities divided among developer, integrator and public customer? Will civil-society participants have access to the same information as industry and governments?

For technology companies, the practical value will depend on whether talks produce shared expectations beyond broad principles: documenting data, testing in conditions close to use, update procedures and incident reporting. Governments must also decide what evidence to require in procurement and how to assess a system throughout its lifecycle. These are issues to document, not obligations already published. [3][4]

At the UN, the current step is consultation. The Forum’s value will depend less on its title than on whether exchanges across sectors produce verifiable practices—without confusing voluntary guidance with law.