Nearly 70% of Websites Sit on an AI-Exposed Flaw

1) Nearly 70% of Websites Sit on an AI-Exposed Flaw

CEOs like Marc Benioff are heeding warnings from Anthropic's Deputy Chief Information Security Officer as the situation escalates.

This week Britain's Office for Budget Responsibility took its website down, as did dozens of other UK government services, Australia's own cybersecurity agency, and government sites in Germany, France, the Netherlands and elsewhere. Their owners pulled the sites as a precaution while installing a fix for two flaws in PHP, the programming language behind WordPress and 69.3% of websites.

The flaws let a stranger take control of the computer that runs a website without logging in, and AI is what helped find them. Calif, a security firm working with OpenAI, found one using OpenAI's GPT-5.6 Sol. Wiz, the cloud-security company Google owns, separately found both using Google's Gemini 4 Argon and GPT-5.6 Sol. "As expected in 2026, AI did most of the work here," Wiz researcher Sagi Tzadik posted. The bugs themselves are old, present in PHP versions from 2012, and a PHP developer spotted one in 2024, but no one found a way to exploit it, and the fix stalled.

Details of the fix leaked a day early, partly through a video of its exploit that Calif posted by mistake, and Cloudflare's CEO counted millions of attack attempts within hours of the release. 

For a site whose owner can install PHP's October 21 update, the fix is in hand, and firewall services like Cloudflare's are filtering attack traffic in the meantime, though Cloudflare says attackers are tweaking their attacks to slip past. The slow part is older sites: by W3Techs data, about 48% of PHP sites run version 8.1 or older, which PHP no longer updates, so they depend on patches from their hosting vendors or a move to a newer version of PHP.

OpenAI, which led a coalition of PHP maintainers and security firms on the response, committed $1 billion in subsidized access to its AI for security work. As AI capabilities continue to advance, Sam Altman said AI models will “probably find bugs… outside of the normal threat model of what we think a hacker can do, hardware-level insane stuff,” meaning flaws in the chips themselves, which software updates often can’t fully fix. “I no longer think we’re going to be able to just patch all the bugs,” he said, arguing that defenders will need to develop “a new kind of defense shield.”

Separately, Microsoft documented a group aligned with the Chinese state using an open-source AI hacking tool to automate attacks targeting hundreds of organizations. Microsoft believes the same group was behind the recent breach of ConnectWise, which makes IT management and remote-access software. 

To understand the stages of attack, the key vulnerabilities, and the larger ecosystem, read our Deep Dive on Cyberwarfare.

2) DeepSeek Raises $15 Billion as Federal Agencies Drop It

On October 20, the Department of Homeland Security published an order barring civilian federal agencies from buying DeepSeek's products and requiring them to remove those products along with any "third-party products and services developed using DeepSeek implementations." DHS cites DeepSeek's servers in China, which Chinese law lets the government access, and the possibility that the models "may contain malware or other security vulnerabilities," and it will review the order within a year.

Two days later, DeepSeek raised about 100 billion yuan (~$15 billion), at a pre-money valuation of 500 billion yuan (~$75 billion). Reuters counts more than 15 investors, with battery maker CATL and Tencent among the largest.

Their updated model, V4.1-Flash-1017, scored 45 on Artificial Analysis’ intelligence index, against 52 to 58 for the top US models. Vals AI, another tester, says that averaged over three runs, it beat OpenAI's GPT-6 Sol.

On October 20 it also released DeepInfer, the core of the software that runs its own inference service for V4.1-Flash on Nvidia and Huawei’s chips. DeepSeek says it shared the code to help others lower compute costs, reduce energy use, and make AI more affordable. In a couple of days, it got over 6 million views and 20 thousand stars on GitHub.

3) Google and Amazon Lower the Cost of Switching AI Chips

This week, Amazon and Google each released a free tool aimed at one of the things that keeps AI companies on Nvidia's chips: rewriting their code to run on another company’s chips, such as Google’s or Amazon’s own. Amazon says 85% of AI researchers and developers write in PyTorch, a free programming toolkit that began at Meta. On Nvidia's chips, PyTorch runs through Nvidia's own software, CUDA, which Nvidia says more than 7.5 million developers use along with its other tools. Moving that code to another company's chips has been slow. 

"You often had to rewrite your code, revalidate it, and sink many weeks in," Amazon CEO Andy Jassy

Amazon's TorchNeuron lets PyTorch code run on Amazon's own Trainium chips with a single configuration change and no rewrite. Google says its TorchTPU, built with Meta, does the same for Google's own AI chips with minimal code changes. Google tried this before, in 2020, but that version required intrusive source-code modifications. In an early test with stand-in model settings, 98% of 4,917 popular AI models ran on TorchTPU without crashing. The speech company Deepgram says its own benchmark found Amazon's chips handled about 60% more audio per dollar than its previous GPU setup.

If these tools make rival chips easier to try and cheaper to run, buyers could gain bargaining power over Nvidia's prices. But both tools are early. Amazon's is a test version, and both companies mostly rent their chips out through their own cloud services. Google itself plans to offer as many as 960,000 of Nvidia's newest chips in its cloud.