Researchers from Wiz, OpenAI, and Calif, some of whom used AI tools including Gemini 4 Argon and GPT-5.6 Sol, discovered vulnerabilities that allow remote code execution and memory access. According to WIRED, these flaws affect nearly all PHP installations serving web requests, including legacy versions. Newsweek reports the vulnerabilities affect PHP versions 5.4 through 8.5.11, and TechCrunch states they have been present since 2004. TechCrunch further reports that the flaws allow attackers to infiltrate server file systems or databases, while ABC News reports they could allow unauthenticated remote code execution. Business Insider reports the flaws were identified as part of OpenAI's 'Patch the Planet' initiative and potentially put 70% of the world's websites at risk of data theft or unauthorized modification. Newsweek lists Joomla as one of the potentially affected major platforms. Sansec has warned of immediate, mass exploitation risks, and Wiz warns that AI makes these previously difficult flaws easier to exploit. In response, WIRED reports that the Port of Rotterdam and several government websites in France and Finland were taken offline as a precaution; Newsweek reports the Wikimedia Foundation took systems offline as a precaution; and the BBC reports the Netherlands' parliament and the BBC took their websites offline as a precaution. Mother Jones reports government sites in Canada were temporarily shut down. The FBI and national cybersecurity agencies in Australia have urged organizations to patch immediately or isolate affected systems, per WIRED. The National Security Administration's cyber division advised web providers to take systems offline or update software immediately, according to Business Insider. The PHP team stated that updating the code is the only complete solution to the risk, and TechCrunch reports the current recommended version for security is PHP 8.5.12, while the BBC reports users of PHP 8.1 or older are urged to upgrade. Additionally, Business Insider reports a Microsoft engineer warned that newer AI models will likely uncover more vulnerabilities in the future, and Mother Jones reports that OpenAI and Greg Brockman suggest the volume of AI-discovered vulnerabilities is increasing sharply, potentially requiring governments to take a larger role in infrastructure monitoring. The Independent reports that Cloudflare implemented stopgap firewall rules to mitigate early attacks, GoDaddy warned that the flaws could lead to data theft or full system compromise, and Wordfence noted that automatic WordPress updates would not fix the issue because the PHP server itself requires updating. The Washington Examiner, citing the Wall Street Journal, reports that the nonprofit foundation managing PHP described the vulnerability as the most severe security flaw the software has faced in over a decade. CNBC reports that Amazon detected exploit attempts before public disclosure via its MadPot system and that OpenAI is leading a coalition to coordinate fixes for the vulnerabilities. WIRED reports that AWS research revealed exploitation attempts began as early as October 17, four days before public disclosure, with AWS detecting early attempts via honeypots and noting the actor used a commercial anonymization network. WIRED further reports that proof-of-concept exploits emerged rapidly, with some appearing hours before the official announcement, and AWS CISO CJ Moses stated that AI tools accelerating vulnerability analysis are shrinking the window between a patch and a working exploit.
The decision by government websites in France and Finland and the Port of Rotterdam to go offline confirms that the risk has moved beyond legacy servers to critical national infrastructure. This validates the systemic danger identified in the first analysis, as the vulnerabilities affect nearly all PHP installations regardless of version.
WIRED provided essential context by revealing that exploitation attempts began four days before public disclosure. This evidence, combined with the use of AI tools like Gemini 4 Argon and GPT-5.6 Sol for discovery, suggests that the traditional embargo and patching cycle is becoming obsolete. The window between a vulnerability's discovery and its active exploitation is shrinking faster than administrators can react.
Mother Jones expanded the scope of the crisis by reporting that the volume of AI-discovered vulnerabilities is increasing sharply. This shifts the event from a failure of software maintenance to a broader question of whether governments must take a larger role in infrastructure monitoring to maintain global digital stability.
While the FBI and national agencies urge immediate patching, the reality remains that legacy versions are fundamentally unpatchable. Current coverage focuses on the immediate urgency of the crisis but continues to omit a deeper investigation into why critical infrastructure still relies on software that cannot be secured.
Source facts (35)
The vulnerabilities allow remote code execution and memory access.
WIRED
The flaws affect nearly all PHP installations serving web requests, including legacy versions.
WIRED
Sansec has warned of immediate, mass exploitation risks.
WIRED
Several government websites in Finland and France, as well as the Port of Rotterdam, were taken offline as a precaution.
WIRED
The vulnerabilities were discovered by researchers from Wiz, OpenAI, and Calif, some of whom utilized AI tools.
WIRED
National cybersecurity agencies in Australia and the FBI have urged organizations to patch immediately or isolate affected systems.
WIRED
The vulnerabilities potentially put 70% of the world's websites at risk of data theft or unauthorized modification.
Business Insider
The National Security Administration's cyber division advised web providers to take systems offline or update software immediately.
Business Insider
The flaws were identified as part of OpenAI's 'Patch the Planet' initiative.
Business Insider
A Microsoft engineer warned that newer AI models will likely uncover more vulnerabilities in the future.
Business Insider
The vulnerabilities could allow unauthenticated remote code execution.
ABC News
The PHP team stated that updating the code is the only complete solution to the risk.
ABC News
The vulnerabilities affect PHP versions 5.4 through 8.5.11.
Newsweek
The AI models Gemini 4 Argon and GPT-5.6 Sol were used to discover the vulnerabilities.
TechCrunch
Joomla is listed as one of the potentially affected major platforms.
Newsweek
The Wikimedia Foundation took systems offline as a precaution.
Newsweek
The vulnerabilities have been present since 2004.
TechCrunch
The flaws allow attackers to infiltrate server file systems or databases.
TechCrunch
Wiz warns that AI makes these previously difficult flaws easier to exploit.
TechCrunch
The current recommended version for security is PHP 8.5.12.
TechCrunch
The Netherlands' parliament and the BBC took their websites offline as a precaution.
BBC
Users of PHP 8.1 or older are urged to upgrade.
BBC
Government sites in Canada were temporarily shut down due to the vulnerabilities.
Mother Jones
OpenAI and Greg Brockman suggest that the volume of AI-discovered vulnerabilities is increasing sharply, potentially requiring governments to take a larger role in infrastructure monitoring.
Mother Jones
Cloudflare implemented stopgap firewall rules to mitigate early attacks.
The Independent
GoDaddy warned that the flaws could lead to data theft or full system compromise.
The Independent
Wordfence noted that automatic WordPress updates would not fix the issue because the PHP server itself requires updating.
The Independent
The nonprofit foundation managing PHP described the vulnerability as the most severe security flaw the software has faced in over a decade.
Washington Examiner
The Washington Examiner reported that the information was originally reported by the Wall Street Journal.
Washington Examiner
Amazon reported detecting exploit attempts before public disclosure via its MadPot system.
CNBC
OpenAI is leading a coalition to coordinate fixes for the vulnerabilities.
CNBC
Amazon Web Services (AWS) research revealed that exploitation attempts began as early as October 17, four days before public disclosure.
WIRED
AWS detected early exploitation attempts via honeypots and noted the actor used a commercial anonymization network.
WIRED
Proof-of-concept exploits emerged rapidly, with some appearing hours before the official announcement.
WIRED
AWS CISO CJ Moses stated that the window between a patch and a working exploit is shrinking due to AI tools that accelerate vulnerability analysis.
WIRED