LatchBio Assesses Grok 4.6 on Biosecurity Tasks

xAI shared findings from an evaluation conducted by LatchBio on Grok 4.6. The review examined how the model handles biosecurity monitoring and adversarial biological tasks. LatchBio reported that Grok 4.6 detects and refuses dangerous queries, even when they are maliciously obfuscated, more reliably than other frontier systems. At the same time the model continues to respond to beneficial scientific queries. xAI pointed readers to a detailed blog post on the LatchBio site and their own biosafety page for the full results of the assessment.