"We used AI to polish sentences" is a bad disclosure, economist argues
paulnovosad · x · 2026-09-24
In a follow-up to his AI-content detection threshold proposal, Paul Novosad adds:
- Permitted boundaries for AI content are imaginable — e.g. AI-generated exhibit notes — provided there's clear disclosure that AI is doing exactly that.
- But "we used AI to refine sentences here and there" isn't a good disclosure: vague labeling erodes the line and invites deception and abuse.
- Together with the earlier "300-word 100% AI" threshold idea, the thread sketches a governance framework: strict detection standards plus explicit disclosure.
More from Safety
- AI Hackers at ~$25 Per Target: Joshua Saxe Warns Security Is Sleeping on Catastrophic Risks — joshua_saxe · 2026-09-24
- AI Now: basic security protocols would have prevented the OpenAI/Hugging Face incident — AINowInstitute · 2026-09-24
- Guardian: AI Overviews reshape search as media outlets sue Google over lost traffic — nordicinst · 2026-09-24
- Dario x-risk-pills the UN Security Council while touting Anthropic's safety record — teortaxesTex · 2026-09-24
- Using eBPF to contain misbehaving AI agents: kernel-level network sandboxing — sloppenheimer · 2026-09-24
- Krueger's bill breakdown: federal AI department, superintelligence ban, 10^25 FLOP pause — DavidSKrueger · 2026-09-24