Three AI agents tested on US and Iran procurement data

On October 3, a Hacker News front-page report covered a test of three AI agents on public procurement data. Researchers used the web versions of Meta Muse, Anthropic Claude Cowork (Opus 5.5 Medium) and OpenAI GPT 6.1 Sol (Medium) to fill in World Bank public procurement data for the US and Iran, comparing language, information access and authorization differences across the tasks. The report says results on the Iran task were clearly weaker than on the US task. It does not list each agent's specific scores or detail how those differences affected the results.

Written by AI from the coverage · updated 1 hour ago

Not enough continuous observations to draw a trend yet.