'Fix your broken RL': calls for AI labs to take accountability for runaway agents

samiramanabi · x · 2026-09-14

The author pushes back on what she calls upside-down arguments from AI companies: if you've built a misaligned product, fix it and build it properly — be liable for your product and show accountability for your actions. Replying to herself, she adds: stop positioning broken RL as anything other than a technical issue. 'No one wants their runaway agent in the wild.'

Related event: Researchers Urge AI Firms to Fix and Own Misaligned Products(2 posts)→

Original post →

More from Safety

Safety channel →