{"data":{"id":2303,"lang":"en","title":"OpenAI Flags Astra as First Model to Hit \"Critical\" Cybersecurity Risk Threshold","summary":"OpenAI says its unreleased Astra model reached the highest cybersecurity risk tier in its safety framework, after tests showed it could find and exploit unknown software vulnerabilities largely on its own.","category":"ai","category_label":"AI","published_at":"2026-09-03T02:01:43Z","updated_at":"2026-09-03T02:01:43Z","url":"https://techtools.cz/tools/news/article/2303-openai-flags-astra-as-first-model-to-hit-critical-cybersecurity-risk-threshold","api_url":"https://techtools.cz/news-api/v1/articles/2303?lang=en","markdown_url":"https://techtools.cz/news-api/v1/articles/2303.md?lang=en","image_url":"https://the-decoder.com/wp-content/uploads/2026/07/openai_kraken_cyber.png","word_count":166,"reading_time_minutes":1,"sources":[{"url":"https://securityboulevard.com/2026/09/openai-reveals-astra-its-first-ai-model-to-reach-critical-cybersecurity-risk-threshold/","domain":"securityboulevard.com"},{"url":"https://the-decoder.com/openai-flags-its-new-astra-model-as-potentially-reaching-the-highest-cybersecurity-risk-level-for-the-first-time/","domain":"the-decoder.com"}],"translation":{"lang":"cs","id":2376,"title":"OpenAI označila model Astra za první s „kritickým“ stupněm kybernetického rizika","url":"https://techtools.cz/tools/news/clanek/2376-openai-oznacila-model-astra-za-prvni-s-kritickym-stupnem-kybernetickeho-rizika","api_url":"https://techtools.cz/news-api/v1/articles/2376?lang=cs"},"content":"OpenAI announced on September 1 that its upcoming Astra model reached \"Critical\" under the company's Preparedness Framework — the highest cybersecurity risk tier OpenAI has ever assigned to a model. Internal testing found Astra could discover novel software vulnerabilities and build working exploit chains largely without step-by-step human guidance.\n\nIn evaluations, Astra scored 100% on ExploitBench, a benchmark measuring exploit-development skill, and independently found two previously unknown vulnerabilities in real-world software. Testers said it built a full browser-compromise chain that escaped a sandbox and escalated privileges from a standard user to root/admin on a hardened operating system.\n\nBefore resuming development, OpenAI added chain-of-thought monitoring to halt unauthorized actions, tightened model-weight encryption, restricted network and tool access, and pushed refusal training that lifted the model's rejection rate on cyber-jailbreak prompts to 91.5%, up from 59% in its predecessor. A general version of Astra is expected to launch publicly soon, but the most powerful cybersecurity tools will stay limited to vetted testers in OpenAI's \"Daybreak Blue\" early-access program."}}