OpenAI admits text watermarks are fragile — detector restricted to approved researchers
OpenAI · x · 2026-10-06
OpenAI acknowledges the limits of its text watermarking:
- Watermarks are often undetectable, especially in short passages; rewriting or translating text can remove them entirely
- For now, only approved researchers get access to the detector to help evaluate and improve the technology
- Text watermarking is treated as an ongoing research area, with continued testing based on feedback from users, developers, policymakers, and researchers
More from Safety
- Data governance giant Collibra acquires Munich-based AI governance startup trail ML — sarahdrinkwater · 2026-10-06
- Fake 'TechCrunch journalist' phishing via X DMs nearly tricks AI researcher — RosieCampbell · 2026-10-06
- Utah becomes first US state to let AI prescribe some medications without doctor review — Polymarket · 2026-10-06
- MCPilot: open-source safety layer for agents to safely tap 38,000 MCP servers — ExcitingHelicopter33 · 2026-10-06
- Cloudflare open-sources six-phase security audit skill for coding agents, 24.7k stars — tom_doerr · 2026-10-06
- AgentHopper: a cross-agent 'AI virus' built on chained prompt injection — wunderwuzzi23 · 2026-10-06