Noisy
← this week  ·  Subscribe →

The world is being quietly rearranged by people who write very long documents.

arXiv · March 30, 2026
The pretty good title  they went with The Competence Shadow: Theory and Bounds of AI Assistance in Safety Engineering Noisy translates that to

AI safety tools are causing engineers to stop thinking, and the math now proves it makes things worse

Adding a second expert to the analysis removes entire categories of hazards from consideration.
What happened
New research shows that using AI to help with safety checks can make human experts miss more problems, not fewer. This means companies cannot just buy AI tools and expect safer systems; they must redesign how people work with the AI.
Why it matters
For years, companies assumed adding AI to safety checks would make things safer. This paper shows that AI can actually make human experts blind to new risks. It means simply buying an AI tool for safety is not enough; companies must now prove their entire human-AI workflow is safe. This shifts the burden from qualifying the software to qualifying the whole process.
The signal
Expect UNECE's GRVA working group and ISO/PAS 8800 revision teams to start fielding requests for workflow-level audit criteria within 18 months, once a regulator cites this paper after the next high-profile autonomous vehicle incident.
If you insist
Read the original →
The Sendoff
The paper warns that AI creates blind spots engineers miss until accidents happen. The paper was written with AI assistance.