Artificial Intelligence

OpenAI's hacker models need proof, not promises

OpenAI says GPT-6.2 Astra and Sol are cheaper and stronger, but cyber safeguards now gate the best security work.

StartupHub.ai Staff

By StartupHub.ai Staff, reviewed by Daniel Singer, Editor

2 min read

OpenAI GPT-6.2 Astra and Sol model announcement
Credit: OpenAI News

OpenAI says GPT-6.2 Astra and GPT-6.2 Sol find and exploit flaws better and write more secure code. The obvious reading is defenders just got faster help.

I am skeptical of that payoff. OpenAI also rates both models at its Critical threshold for cybersecurity.

That rating triggered tighter safeguards for higher-risk help, OpenAI said. The same filters will restrict some legit vulnerability research and testing. Approved defenders keep reduced refusals on Daybreak Blue and GPT-6 Cyber on Daybreak Red, while GPT-6.2 waits for Daybreak access.

OpenAI tested cyber skills with safeguards disabled, per its system card. It names DeepBinary, ExploitBench, SecureSWE and PatchBench but gives no exploit rates in the announcement. For the obvious reading to hold, the patches must preserve behavior at scale and secure-code gains must survive real repos.

To be fair, OpenAI pairs capability with efficiency. It prices Astra at $8 input, $0.40 cached input and $40 output per million tokens, 20% below GPT-6 Astra, per OpenAI. Astra for Law averaged 0.03 material errors per assignment, half the base rate and below Claude Opus 5.5, per OpenAI.

Both models also hit High for biological and chemical capabilities, OpenAI said. Astra is the first external release to hit High for AI self-improvement. If you lead a security team, pilot GPT-6.2 on your own backlog before you change staffing.

Sources

Get the free daily brief every morning

Every morning the day's AI news, new AI startups and new funding rounds. Free and takes a few minutes to read, unsubscribe at any time.

Join 64,000 other subscribers.

TopicsOpenAIGPT-6.2 AstraGPT-6.2 SolHarveyClaude Fable 5.2DaybreakAstra for Law

StartupHub.ai StaffDaniel SingerWritten by StartupHub.ai Staff, Editorial teamReviewed by Daniel Singer, EditorAbout

StartupHub.ai Staff, Editorial team

The staff writers of StartupHub.ai, ranging from investment analysts to avid AI tool users, early adopters and critical enthusiasts. Backgrounds span engineering, business and the arts. We hold every piece to rigorous standards of research and review.

Daniel Singer, Editor

Daniel Singer is an AI expert who has been writing about, covering and building AI technology since 2018.

More from Daniel Singer

Test your Agent Readiness now

Paste your website. How well ChatGPT, Claude and the other agents can read and use a site, in about a minute.

Watch

The videos behind the news, one flip at a time

The feed mixes the videos from our articles with startup demos and you flip through them like TV channels for free with no signup.