International governments and agencies have urged websites to address two critical cyber-security flaws in the PHP programming language, which powers a significant portion of the internet. A fix was issued on Wednesday afternoon, but multiple agencies advised organizations unable to apply it in time to consider taking their sites offline. This led to temporary outages for entities such as the Finnish government, Edinburgh council, and some BBC websites.

The vulnerabilities were discovered using advanced artificial intelligence, with cyber-security firm Wiz describing them as the most severe flaws ever found by AI. These vulnerabilities can reportedly provide a direct path to a server's file system or database. Finland has temporarily shut down its public web services, and the Netherlands' cyber-security agency has taken its own website offline as a precautionary measure.

The US National Security Agency issued advice for organizations to take systems offline if immediate updates were not possible. The White House stated that federal agencies are collaborating with leading AI labs to expedite a resolution. Japan's Prime Minister, Sanae Takaichi, also urged checks for publicly available websites running on PHP.

The Wikimedia Foundation, responsible for Wikipedia, confirmed it was addressing the vulnerabilities and had taken down some systems as a precaution, resulting in temporary access issues for parts of Wikipedia. Edinburgh council's website and online platforms for several BBC programs, including Doctor Who and The Traitors, also became inaccessible, though a direct link to the PHP flaws has not been confirmed for these specific cases.

Other platforms, like the art site DeviantArt, have announced temporary maintenance without specifying the cause. The flaws were identified by researchers at Calif and Wiz, who participated in OpenAI's "Patch the Planet" program. This initiative provided them access to advanced AI tools to find and fix web development language issues.

OpenAI noted that the discovery represents one of the most significant security issues in recent times, with an internal researcher also identifying the problem around the same time as the external firms. These flaws had remained undetected for over a decade before being revealed through AI analysis over a few weeks in August.

Wiz stated that frontier AI models were crucial in developing and adapting exploits across multiple applications rapidly. The firm estimates that approximately 70% of the internet could be affected, a figure the BBC has not independently verified. The PHP Foundation had planned to release its findings and a comprehensive fix later this month.

While the immediate focus is on applying the security update, the incident highlights the growing role of AI in both discovering and potentially exploiting cyber vulnerabilities. The speed at which these flaws were found and the potential scale of their impact underscore the evolving challenges in maintaining global cyber security.