← Accueil actu | Salle du marché
2026-09-10 · Source：InfoWorld
OpenAI's new flagship scored 100% on ExploitBench without safeguards, triggering deployment restrictions as it rolls out to ChatGPT tiers, the API, and Amazon Bedrock.
OpenAI released GPT-6 Astra on Thursday, marking the first time one of its models has crossed the "Critical" cybersecurity risk threshold under the company's internal Preparedness Framework. That classification, the company said, activates additional deployment restrictions. Astra is initially available to a limited set of organizations, and over the coming days it will reach all ChatGPT Plus, Pro, Business, and Enterprise subscribers, as well as customers of the OpenAI API and AWS.
Deployment is deliberately gated. Enterprise administrators must manually switch on Astra for their workspace, because access is disabled by default at launch. Developers can call the model in the API under the identifier gpt-6-astra or run it through Amazon Bedrock. Pricing is set at $10 per million input tokens and $50 per million output tokens. Subscribers on Pro, Business, and Enterprise tiers also receive a variant named Astra Pro, and OpenAI said Astra supports Zero Data Retention for eligible API customers.
The benchmark results underpinning the Critical rating are striking. OpenAI said it evaluated Astra with production safeguards removed on ExploitBench, where the model achieved a perfect 100% score, up from 78.5% for its predecessor, GPT-5.6 Sol. On ExploitGym, a broader measure of exploit-development capability, Astra posted a 42.4% success rate compared with 30.3% for Sol — and did so while consuming fewer output tokens.
In the company's framing, the same capability that makes Astra a security concern also makes it a defensive tool. "Its ability to identify and develop zero-day exploits can help defenders find and patch weaknesses, but it also creates a need for stronger safeguards," OpenAI said. The combination of a top-tier risk rating and a staged, opt-in rollout illustrates the deployment controls OpenAI is applying to models that cross its highest internal risk classification.
Mots-clés: OpenAI · GPT-6 Astra · ChatGPT · Amazon Bedrock · Cybersecurity · AI Safety
萬安算交所 AIXX · AIXX · Bourse compute vérifiée · Accéder à la salle