Threat Brief — 2026-09-07 — Astra Goes Wide
OpenAI has begun rolling out ChatGPT Astra — its most capable model, previously reported to score 100% on ExploitBench for zero-day discovery and exploit writing — to $20/month Plus subscribers. This dramatically expands the population with access to a model demonstrated to automate vulnerability research and exploit development. No other fresh findings were ingested in this cycle; the broader threat landscape remains consistent with recent days.
Top items
- ChatGPT Astra now available to Plus subscribers. OpenAI is expanding access to Astra beyond its initial launch tier to the $20/month Plus plan, though free users remain excluded. Astra was first reported on 2026-09-02 as scoring 100% on ExploitBench, raising concerns about automated zero-day discovery and exploit generation at scale. Widening access to this model increases the pool of individuals who can leverage AI-assisted offensive capabilities. (src: BleepingComputer) — Developing story: first reported 2026-09-02 by SecurityLab.ru.
Themes
AI capability diffusion. The move from restricted launch to broad commercial availability of a model proven on ExploitBench marks a significant step in normalising AI-assisted offensive tooling. Combined with recent reporting on live AI-assisted attacks by multiple threat groups and warnings about a narrowing window before autonomous AI-driven attacks become routine, the accessibility trend bears watching even though no immediate attack activity is linked to today's rollout.