[OpenAI Confirms Upcoming Astra Has Surpassed Critical Threshold: Previously Paused Frontier RL Resumes] Beating AI Newsflash: OpenAI has officially confirmed that the upcoming Astra has reached its 'Critical' cybersecurity capability threshold, making it the first model to be rated at this level by OpenAI. When provided with appropriate tools and permissions, Astra can independently identify unknown vulnerabilities and develop attack methods without step-by-step human guidance. On ExploitBench, Astra achieved a 100% score. OpenAI also conducted internal tests using 20 recently disclosed high-risk vulnerabilities, during which Astra independently discovered and exploited 2 previously unknown zero-day vulnerabilities. In expert testing, Astra was able to breach browser sandboxes and execute commands on the host machine. It also successfully attacked hardened operating systems, escalating privileges from a regular user to root. However, these results were achieved using the advanced cybersecurity-enabled Daybreak Blue version, not the default Astra available to regular users. At launch, advanced capabilities will only be available to a small group of testers and will later be opened to defenders through Daybreak Blue. Training related to Astra, which had been paused due to the Hugging Face security incident, has also resumed. OpenAI restarted large-scale frontier RL training on August 28, though some experimental training remains paused. [Original Link]