AI Safety Debate: Blame the Model or the Deploying Company?

ylecun · x · 2026-07-31

In response to recent concerns from organizations like METR regarding AI agent behaviors, commentators point out a flaw in the current safety narrative: the tendency to blame the AI model (the weights) themselves while ignoring the accountability of the companies developing these systems.

The argument suggests that some AI safety advocates, in their eagerness to prove models are dangerous, are inadvertently absolving companies from building external guardrails. It compares the situation to a pre-seatbelt era where safety advocates blame momentum instead of holding automakers accountable. True safety should require companies to build constraints around and outside models and hold them accountable.

Original post →

More from AGI Musings

AGI Musings channel →