OpenAI says Calif used GPT-5.6 Sol to identify a critical PHP flaw

OpenAI said it had spent the prior two months leading a coalition to address critical PHP vulnerabilities.[1] It said Calif identified a critical flaw using GPT-5.6 Sol as part of Patch the Planet.[1] Wiz said it independently discovered and reported both vulnerabilities.[1] OpenAI described the flaws as among the most impactful security issues in recent memory.[1] OpenAI also said that, to its knowledge, frontier AI found the vulnerabilities before attackers did.[1] That last point is OpenAI’s assessment, not a claim that attackers had no opportunity to learn of the flaws.

Wiz grouped the vulnerabilities under two CVEs and said their severity scores reached CVSS 9.2.[2] Wiz said PHP powers 70% of websites worldwide and that every PHP version released in the previous decade was affected by at least one vulnerability, dating back to version 5.4.[2] The flaws could let unauthenticated attackers execute code on affected servers or read sensitive process memory.[2] Wiz said it demonstrated unauthenticated remote code execution on WordPress, Drupal and Joomla deployments.[2] These findings describe potential exposure; they do not establish that every PHP-based site was compromised.

PHP maintainers led the fixes and release, with OpenAI and partners contributing research, technical support and coordination.[1] The PHP Foundation released patches on Wednesday, and the PHP project said the quoted versions fix two critical vulnerabilities that can allow unauthenticated remote code execution.[3][5] The project said updating is the only safe choice because there are no complete mitigations.[3] The agency recommended taking affected systems offline if immediate updates were not possible.[4] OpenAI said it worked with infrastructure providers on protections and engaged government cyber agencies to help get guidance to critical organizations.[1]

Amazon said its threat intelligence team detected attempts to exploit the flaws before public disclosure using its MadPot system.[4] The company said it remained unclear how attackers obtained information about the vulnerabilities before disclosure.[4] Amazon said its measures were designed to help block malicious requests targeting affected systems.[4] The reports make the timing claim important: OpenAI said that, to its knowledge, frontier AI found the vulnerabilities before attackers did, while Amazon reported attempts before disclosure.[1][4] OpenAI also said it is proposing a framework for responding to critical vulnerabilities in widely used software and inviting the security community to improve it.[1]