OpenAI publishes a post on pacing model development against cyber-capability thresholds

Sources: OpenAI post, HN discussion, second HN thread
Summary: OpenAI published a post dated 2026-08-18 on pacing model development in an era of cyber-critical capabilities. openai.com returned HTTP 403 to the fetcher for the second consecutive day, so the post was not read and nothing from its text is asserted here. The existence of the post is the only established fact in this block, which is why the status is developing.
Why it matters: A frontier lab gating releases on offensive-security evaluations would move model availability timelines for anyone building on its API.
Follow-up: Read the post once openai.com is reachable and establish whether it states a training decision, and whether any scope, duration, or trigger is named.