# @BronsonSchoen — 2026-04-16

♥10 ↻1 · https://x.com/BronsonSchoen/status/2044630122385518803

I’m surprised it took models this long tbh (I think anthropic models were ahead of the game here, it’s kind of insane some of the older evals still get hits). 

I’ve never understood your theories on evaluation awareness as far as how they’d apply to the trajectory current frontier labs are on though.

As OpenAI notes, models sometimes think this even in literal production.

tags: author:bronsonschoen, kind:tweet, thread-context, year:2026