Evidence item

OpenAI’s safety overview supports the September 3 release and Critical cybersecurity threshold; release timing alone does not independently validate a universal capability ranking. DIRECT_TEXT

Authenticity UNVERIFIED · extraction UNVERIFIED · independence group none (unreviewed) · created seq 8632

Source: Safety overview: GPT-6 Astra · locator {} · excerpt hash sha256:a868920b51e36cb7790b45a9092094f0784eee93ce2f247728b1fe01bc25790e

Exact excerpt

OpenAI released GPT-6 Astra on September 3, 2026, describing it as the first model to meet the Critical cybersecurity threshold under its Preparedness Framework.

Linked claims

Claim	Direction	Relevance	Steps	Counted
bd8259fc	SUPPORT	DIRECT	0	yes

Audits

None.