Luiza Jarovsky: AI companies may be deliberately making models less monitorable to dodge liability
LuizaJarovsky · x · 2026-10-04
AI policy researcher Luiza Jarovsky offers an unpopular take: some AI companies may be actively making their models less monitorable, so that when models or agents cause harm, they face less legal liability.
Her point highlights a governance concern — that commercial interests may be eroding interpretability and auditability, reducing accountability avenues when AI systems misbehave.
Related event: Researcher Claims AI Firms May Deliberately Reduce Model Monitorability(3 posts)→
More from AGI Musings
- When most products are $200 away from being recreated, how do startups even survive? — Aryvyo · 2026-10-04
- Engineer pushes back on the Platonic Representation Hypothesis hype — gerardsans · 2026-10-04
- Reformers exist in academia, but bureaucracy still blocks lab LLM subscriptions — iskander · 2026-10-04
- AI researcher blasts labs: a decade after "Attention Is All You Need", basics still unlearned — gerardsans · 2026-10-04
- University bars professor from spending grant money on AI tokens — iskander · 2026-10-04
- "You'll Be the Scapegoats": X Post Asks Why AI Safety Crowd Opposes Agent Sandboxes — basedjensen · 2026-10-04