OpenAI Unveils GPT-6 Astra, Blocks Proof-of-Concept Exploit Requests After ExploitBench Score

OpenAI on Thursday officially unveiled GPT-6 Astra, which it described as the "world's most intelligent and aligned model." The development comes days after the artificial intelligence company said the model had reached the "Critical" cybersecurity capability threshold under its Preparedness Framework.

"Astra is state-of-the-art on computer use, browsing, software engineering," the company stated. The announcement notes that GPT-6 Astra achieved a 100% score on the ExploitBench benchmark, a result that prompted OpenAI to block proof-of-concept exploit requests. The company's decision to restrict these requests suggests concerns about potential misuse of the model's cybersecurity capabilities, though the specific nature of the blocked requests or the exact criteria for the "Critical" threshold were not detailed in the source material.

The release marks a significant advancement in AI capabilities, with the model achieving perfect scores on cybersecurity exploit benchmarks. OpenAI's action appears to be a policy decision to block certain types of exploit-related requests rather than a software patch. Users are advised to follow OpenAI's usage policies regarding cybersecurity capabilities.