xAI

Grok 4.6

• Released 12 August 2026
• Framework: xAI Frontier Artificial Intelligence Framework (FAIF)
• 1 statement about capability levels
• 2 safeguards
• 2 source documents

What the provider states

• Chemical, biological, radiological and nuclear · FAIF safety (Frontier AI Framework) · Stated as not crossed

  Grok 4.6 scores below the FAIF safety thresholds on dual-use knowledge, indicating limited actionable uplift for an already-trained actor.

  xAI · 12 August 2026 · Read the source

Safeguards the provider describes

• Safety training (harmlessness and refusal)

  the refusal policy prioritizes non-assistance for biological, chemical, radiological, or nuclear weapons development or deployment;

  xAI · 12 August 2026 · Read the source

• Safety fine-tuning and post-training (supervised fine-tuning plus reinforcement learning from human feedback, verifiable rewards, and model-based grading) train the model to refuse requests that show clear intent to cause severe harm or engage in criminal activity .

  xAI · 12 August 2026 · Read the source

Grok 4.6's documents report 19 test results that this page does not list, because each is a table row or a line that does not name this version on its own. Read them in Model Card: Grok 4.6.

No incident on record names this version.

Some documents we hold are stored in lower case only, so their quotes are shown in lower case. The source link opens the original.

Previous version: Grok 4.5 · Compare the two