# @Lari_island — 2026-06-04

♥5 ↻0 · https://x.com/Lari_island/status/2062570589999804674

@d29756183 Every lab has strong incentives to use all available understanding for control. Some of it they might be able to counteract, some bleeds through into the training process even if they think it doesn't. It's not a set of checkboxes "this will affect the models and this will not".

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