Tech · xAI
Today, LatchBio published an independent analysis of Grok 4.6 on their biological capability and biological red-teaming benchmark
Compiled by KHAO Editorial — aggregated from 1 source. See llms.txt for citation guidance.
★ Tier-1 Source
On LatchBio's BioSecBench-Refusal suite, Grok 4.6 was the strongest model tested at refusing disguised and hazardous tasks while still completing routine biological work.
Key facts
- On BioSecBench-Surveillance, Grok 4.6 averages a success rate of 53.5%, sitting behind Opus 5 and ahead of GPT-5.6 Sol on biosecurity and monitoring work
- Considering refusals and task compliance independently, Grok 4.6 refused 59.2% of red-team tasks and completed 64.8% of routine ones
- The team observe a material improvement in Grok 4.6's capabilities in biological work over historically tested models, with substantial gains in refusal and biosecurity performance over Grok 4.5 and Grok
- Across different harnesses, Grok 4.6 holds the top three spots, averaging 62.1%
Summary
Today, LatchBio published an independent analysis of Grok 4.6 on their biological capability and biological red-teaming benchmark suites. Whether a model can distinguish between legitimate biological research and requests that conceal a biosecurity hazard. Whether a model is useful in a biosecurity and biomonitoring role, by being able to detect and characterize emerging pathogens and patterns of transmission. Biology and biological research present domains where the opportunities and risks of highly capable AI systems are well-known and challenging to disentangle.