Skip to main content

Publication

Back to Resources

Gender and WPS in military AI: implications for IHL compliance

26 November, 2026

Authors

Description

This policy brief examines how AI-enabled military systems, from intelligence analysis and targeting to operational planning and civilian harm assessment, affect states’ compliance with international humanitarian law (IHL). It argues that gender analysis and the Women, Peace and Security (WPS) framework are operational tools, not peripheral concerns. Datasets and models can reproduce social biases, which can distort assessments of who is present in an area, how an attack will affect different civilians, and what precautions are needed to minimize civilian harm. Each of these risks bears directly on the IHL principles of distinction, proportionality and precautions in attack.

Gender bias and data limitations (…) must be understood not only as technical challenges but as a potential source of legal risk.

Drawing on an expert roundtable held in March 2026 by DCAF and the Canadian Red Cross, the brief sets out six recommendations for states across the AI life cycle. They cover treating gender bias and data limitations as legal risks; establishing gender-responsive governance and oversight; using AI to strengthen civilian harm mitigation; embedding gender considerations in Article 36 legal reviews and in procurement; and strengthening women’s meaningful participation and gender expertise in AI governance. The brief also addresses automation bias and the limits of human judgement and control in high-tempo operations, and notes that use may need to be restricted where bias cannot be mitigated.

The brief is intended for military legal advisers, defence policymakers, operational planners, procurement authorities and others involved in governing AI-enabled military systems. Its recommendations can inform doctrine, legal review, procurement specifications, training and oversight, and they apply before systems are fielded, when the decisions that shape compliance are made.