Study shows readers cannot identify watermarked AI text
tokenbender · x · 2026-08-22
An experiment tested whether readers could distinguish between AI text watermarked with SynthID-Text and unmarked text.
Results:
- Random Chance: A purely random selection yields an average score of 3.33/10.
- Actual Performance: After correcting for order bias, the mean score of 278 participants was 3.4/10, statistically indistinguishable from random guessing.
- Conclusion: Current AI watermarking is effectively invisible to the average reader and cannot be reliably identified.
More from Safety
- Anthropic red team's first-ever model "abliteration" cut GLM-5.3 refusals from 90% to 3% for ~$4,400 — cephaloform · 2026-10-16
- Anthropic grants METR employee-like access as independent safety evaluators go public — AnthropicAI · 2026-10-16
- APEX-Harm benchmark: script-based injection attacks easily compromise agents in shared workspaces — KAUST · 2026-10-16
- Residual-stream forecasts predict agent violations before they happen, USC study shows — USC · 2026-10-16
- AI safety drama: risk research called a 'company effort,' team lead responds — logangraham · 2026-10-16
- Brave's Leo adds end-to-end encrypted chat with GLM via attested TEE — gnukeith · 2026-10-16