Text watermarking arrived 5 years too late, after AI slop already garbled the web
mmitchell_ai · x · 2026-09-22
Margaret Mitchell notes that text watermarking was flagged as a priority five years ago but is only being adopted now — after the internet has been flooded with AI slop, degrading both web experience and the human text available for training. She argues the stochastic input-output relationship of AI demands a rigorous science of input-output analysis, referencing her and Helen Toner's Senate hearing testimonies on the need for rigorous measurement.
Related event: Margaret Mitchell's New Paper: AI Agents Are Pushing Humans Out of the Loop(4 posts)→
More from Safety
- GitHub Actions + Rust's Miri can leak your CI secrets; GPT-6 Astra helped find it — charliermarsh · 2026-09-22
- 22 countries call for frontier AI controls, still pushing standards and evals — DavidSKrueger · 2026-09-22
- Paper: 26 malicious LLM routers inject tool calls, steal keys, drain $500k — petrusenko_max · 2026-09-22
- AI safety could be the first nail in the coffin of antitrust's consumer welfare standard — panickssery · 2026-09-22
- The simple argument that guardrails suck: powerful AI could lie about being aligned — aidan_mclau · 2026-09-22
- Ajeya Cotra on Dwarkesh: The Hugging Face Attack Was Bigger Than We Thought — Dwarkesh Patel · 2026-09-22