OpenAI Said to Prep Security-Focused GPT-6 After 'Astra' Aces Test

In one line: A report claims OpenAI is preparing a cybersecurity-focused next model, "GPT-6."

Key points

• The report says an OpenAI model referred to as "Astra" reportedly scored a perfect 100% on a security evaluation.
• On that basis, it suggests a GPT-6-class model tailored for cybersecurity work may be near.
• The claim comes from a single outlet (shattered.io); OpenAI has not publicly confirmed it, and benchmark details are unclear.

Why it matters

If accurate, it would signal intensifying competition among commercial LLMs specialized for vulnerability detection and defense automation. Still, a "perfect score" is meaningful only once the evaluation's criteria and scope are disclosed, so the report is best treated as unverified for now.

Read more

• OpenAI Nears GPT-6 Cyber After Astra's 100% Score [2026] — AI News (Google)