Test shows DeepMind's SynthIDBio protein watermark can be washed out, researchers say
owl_posting · x · 2026-10-03
In a quick test with Raygun, researchers found that DeepMind's SynthIDBio protein sequence watermark signal could be washed out while preserving the predicted protein structure.
They argue protein sequence watermarks buy little real biosecurity, writing that "biosecurity is too important for security theater."
More from Safety
- "Conscious AI" claims should be treated as an AI safety issue, argues Jarovsky — LuizaJarovsky · 2026-10-04
- Miles Brundage: shipping with known flaws isn't iterative deployment, it's just faster shipping — Miles_Brundage · 2026-10-03
- OpenAI's new reasoning models can have chats human-reviewed even if you opted out — JeremyNguyenPhD · 2026-10-03
- NUS Press Unveils a 'Very Sensible' AI Writing Policy, Economist Endorses Its Rationale — paulnovosad · 2026-10-03
- Ex-OpenAI safety lead: 12 launches left no time to fix structures, some mistakes may not be iterable — CurieuxExplorer · 2026-10-03
- Trace tampering research covered by Fast Company — maksym_andr · 2026-10-03