Many in AI safety have narrowed in on automated AI R&D as a key risk factor in AI takeover. But I'm concerned that the actions they're taking in response (e.g. publishing evals, raising awareness in labs) are very similar to the actions you'd take to accelerate automated AI R&D.