CLAIM-2026-09-03-04

OpenAI reports a 91.5% refusal rate on cyber jailbreaks at the model layer and a 100% score for Astra on ExploitBench.

Why it matters

Refusal and benchmark rates are safeguard properties of the vendor's configuration; a downstream team cannot reproduce them and should not inherit them as assurances.

Sources

• openai2026path