OpenAI system card: GPT-6.2 Astra and Sol hit Critical cyber threshold, High for bio
OpenAI says both models meet its Critical cybersecurity threshold and its High biological and chemical threshold, and it applies safeguards to internal use and to external monitoring.
What happened
OpenAI published a safety document, called a system card, for two new models: GPT-6.2 Astra and GPT-6.2 Sol. The company says both are strong enough at hacking-related tasks to reach its top cybersecurity risk level, called Critical. For biology and chemistry, it rates them one step lower, at High.
Why it matters to you
Models that can find software flaws or help with dual-use science raise real misuse risks, so the safeguards around them affect anyone who relies on software or online services.
The takeaway
OpenAI says these models are powerful enough to need extra monitoring and tighter refusals.
What it means for you
Expect more refusals on dual-use prompts such as autonomous vulnerability discovery and chained exploitation, per OpenAI.
OpenAI says refusals, blocks and bans protect its General Access products, and a monitoring system also covers external use of the models.
The source reports safety thresholds and benchmark results only, with no financial figures.