22 October 2026
Google, OpenAI and Amazon reveal how AI helped find the PHP flaws, and who struck first
Google and OpenAI say AI models helped uncover the 'PHP2Shell' flaws, Amazon says attackers were probing them before disclosure, and the NSA and FBI issued warnings.
Yesterday Noorult covered the PHP security flaws that pushed government websites offline. The new part of the story is who found the holes, who spotted hackers using them, and how much of that work was done by artificial intelligence.
What's new
The two weaknesses now have a name, PHP2Shell. Researchers say they could let an attacker run their own code on a web server remotely, which could expose private data. Not every site that runs PHP is necessarily at risk.
Amazon said its threat intelligence team caught attempts to exploit the flaws before they were made public. It spotted them with MadPot, a system that watches for suspicious traffic across internet-facing machines. Amazon said it still doesn't know how attackers learned about the bugs early.
Google said its security unit Wiz used AI models to help find and study the flaws, and that its Gemini AI helped judge how serious they were, test possible attacks and check that the patches worked.
OpenAI, which is private, said researchers using its models also helped uncover the bugs. It said it had spent two months leading a group of software maintainers, security researchers and security companies to line up fixes before the public announcement. It ranked the flaws among the most serious security problems in recent memory.
Cloudflare said it had added protections to its web application firewall, a filter that blocks bad requests before they reach a website, as a backstop while owners install updates. PHP's developers have released fixes.
Governments weigh in
In the US, the National Security Agency told organizations using PHP to get updates ready, and said systems that can't be patched right away should be taken offline. It warned the flaws could give attackers direct access to a server's files and databases. The FBI issued its own warning. Cyber agencies in Singapore, the UK, Australia, Canada and the European Union put out similar alerts.
Why it matters
This is one of the clearest public examples yet of AI cutting both ways in cybersecurity. The same kind of model that helped defenders find and fix these bugs could, in the wrong hands, help criminals find the next one. OpenAI made that point itself. For investors, it is also a reminder that some of the biggest cloud and AI companies now sit at the center of the internet's defenses, so how they handle episodes like this one is part of what they sell.
What to watch
- How Amazon and others explain the attacks that started before disclosure.
- Reports of successful break-ins at sites slow to patch.
- Whether more AI companies disclose bugs their models helped find.
The companies in this story
- GOOGL: financials, earnings and dividends
- AMZN: financials, earnings and dividends
- NET: financials, earnings and dividends
Based on reporting by CNBC: Critical web security flaws spark global alarm: Google, OpenAI, Amazon. Facts, not advice.
More news
- Anthropic's IPO, in numbers: 825 million shares at $70 to $78, and a $1.4 trillion price tag
- Trump says the Nvidia scandal didn't endanger US security, and stands by Lutnick
- Nvidia's first earnings without Huang: what Kress will be asked on Tuesday
- Buffett gives away $5.5 billion of Berkshire stock in his Thanksgiving letter
- Banks are using AI to trim back offices. Proof that it adds revenue is harder to find
- IMF chief says the world economy leans too hard on AI as circular deals top $1 trillion