The Problem

The AI industry spends billions every year on evaluation and data tasks. Despite these enormous sums, everything happens behind closed doors. All we have is their word, with no actual proof.

Models are getting released with massive flaws. Developers have been focused on releasing as fast as possible, while red teaming and testing happens inside private labs with no external scrutiny. AI deployments outpace evaluation.

There is currently no accessible path for meaningful public involvement in AI safety. No rewards or incentives for human participation. Even well-intentioned public competitions remain limited to technical developers running for a few weeks.

The people who will be most affected by AI systems have no meaningful way to shape how those systems behave.