Skip to content
MaungaAI news, explained

Safety and security

PHP flaws are now being used to break into servers tied to the US government, Dream says

An attacker, possibly linked to China, reportedly used the PHP flaws found with AI to break into 25 servers, three of them tied to the US government. The report comes from the company Dream, and no one else has confirmed it.

4 min read7 sourcesWritten with AI · machine-translated

Share: WhatsApp · Email

AI-generated illustration for Maunga. Not a photograph.

The essentials

  • Claimed by one partyAccording to the company Dream, an attacker used the PHP flaws to take control of 25 servers, three of them tied to the US government (ciberseguridad).
  • Claimed by one partyDream found Claude Code and keys for other AI models in the attacker's infrastructure, and suspects a group linked to China (ciberseguridad).
  • On October 21, the PHP Foundation released patches for two flaws that let someone take control of a server without a password (TechCrunch, Ars Technica, Simon Willison's Weblog).
  • Claimed by one partyAmazon says the flaws are being widely exploited. On October 24, the PHP Foundation said it had not seen any successful attacks (Ars Technica, WIRED).
  • Claimed by one partyAccording to the companies themselves, Calif and Wiz found the flaws with the help of AI, including the models GPT-5.6 Sol and Gemini 4 Argon (TechCrunch, WIRED, Simon Willison's Weblog).

Why it matters

If Dream's report is confirmed, the PHP flaws no longer threaten only websites. They also threaten devices that run PHP inside, like these firewalls, which can serve as a door into a public agency's network (ciberseguridad). This suggests that updating the website's server isn't enough: you also need to check which other devices use PHP without anyone noticing.

AI shows up on both sides. It helped find flaws that the PHP team partly knew about but thought were too hard to exploit (WIRED). Wiz warns that this barrier is gone, because a model can adapt the attack to each version (TechCrunch). Even so, the experts quoted by WIRED think few flaws will have such a wide reach (WIRED).

What does it mean for me?

If you have your own website, even a WordPress site on cheap hosting, ask your provider whether it has already updated PHP. WordPress says the fix happens on the server, not on the website (TechCrunch). If you just browse the web, you don't need to do anything.

The details

The Israeli company Dream has described an attack it observed on October 25 (ciberseguridad). According to its report, the attacker combined the PHP flaws with another flaw in the Linux kernel, called Fragnesia. This let it take full control of 25 servers in the United States, Japan, Taiwan and South Korea. They were firewalls, the devices that filter a network's traffic, made by a large manufacturer that Dream doesn't name (ciberseguridad).

Three of those devices were in an environment tied to the US government. From one of them, the attacker opened a tunnel into the internal network. It then tried to extract passwords and create hidden accounts, and Dream doesn't know whether it succeeded (ciberseguridad). In the attacker's infrastructure, Dream found Anthropic's Claude Code installed, along with keys for using models from DeepSeek and Kimi. Based on the countries targeted, Dream suspects a group linked to China (ciberseguridad).

The data on attacks doesn't fully add up. Amazon said on October 21 that it had seen an attempt four days before the announcement and that the flaws were already being widely exploited (Ars Technica). According to WIRED, Cloudflare recorded millions of attempts in a few hours. But the PHP Foundation had not seen any successful attack, and many of the attacks going around seemed not to work (WIRED). The note on Dream's report also mentions limited exploitation (ciberseguridad).

According to TechCrunch, the flaws had been in PHP since version 5, from 2004 (TechCrunch). WIRED reports that Calif found them in early August with the help of GPT-5.6 Sol, as part of OpenAI's "Patch the Planet" program. Wiz found them on its own with Gemini 4 Argon and GPT-5.6 Sol (WIRED, Simon Willison's Weblog). The PHP Foundation moved the patch up to October 21 because the identifiers for the flaws leaked on October 20 (Simon Willison's Weblog).

What we don't know

  • Whether the attacker described by Dream obtained valid passwords or hidden accounts on the network tied to the US government. Dream itself doesn't know.
  • Which firewall manufacturer is affected, and whether anyone outside Dream has confirmed its report.
  • How many websites and devices are still not updated.

Background

According to WIRED, some of the underlying problems were already spotted in 2024. At the time, they were considered too complex to exploit and were not fixed (WIRED). Ars Technica notes that this year, new models from Anthropic and OpenAI with cyberattack capabilities led the US government to temporarily limit their rollout (Ars Technica).

How this story has changed

  1. Added the attack described by Dream against 25 servers, three tied to the US government, and WIRED's data on attack attempts and the PHP Foundation's position.
  2. Amazon says the flaws are already being widely exploited and that it saw an attack attempt four days before the announcement; added the British warning, Cloudflare's filters and OpenAI's checking tool.
  3. We published the story.

Sources

This story is based on 7 sources. In the text, each fact names in brackets who reports it.

Was this story useful? Yes Not really

Spanish original