GoKawiilBenchmark finds leading AI models omit Women, Peace and Security considerations
Our Secure Future, using the Collective Intelligence Project's Weval platform, built a public benchmark testing whether large language models apply Women, Peace and Security principles in peacekeeping contexts. The results show models can recognize WPS concepts when prompted but fail to raise them unprompted, a pattern the group calls the WPS Competence Gap. OSF says this gap means AI systems largely ignore the experiences of over half the global population, particularly in the Global South.
GoKawiil's interpretation of the reporting above, not reported fact.
As peacekeeping officials and military intelligence analysts increasingly consult AI tools for decision-making, gaps in gender and conflict awareness could shape real-world judgments without anyone noticing the omission. The benchmark suggests that embedding domain expertise from civil society groups into AI evaluation could surface blind spots that standard testing methods miss. This also illustrates a broader push to use participatory, expert-driven evals to hold AI systems accountable in high-stakes social and political domains.
- OSF and CIP released a public benchmark measuring AI adherence to Women, Peace and Security principles.
- Models reportedly understand WPS concepts but don't volunteer them without direct prompting.
- Peacekeeping and military intelligence professionals are already using LLMs, raising the stakes of this gap.
Source: blog.cip.org, 2026-11-12
Published there as: “AI Models Know What Women, Peace and Security Is. They Just Don't Bring It Up”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.