Anthropic’s AI discovered a flaw in GNU/Linux that allows an outsider to take control of a server over the internet—without needing a password. GNU/Linux powers 70% of websites (where the underlying system can be identified), and this bug had remained hidden in the code for about 10 years. It has a severity score of 9.8 out of 10, and Red Hat has classified it as critical. In Anthropic’s tests, the attack requires about 48 minutes of preparation, followed by another 5 minutes to gain full control of the machine. A fix for GNU/Linux has been available since October; however, as of yesterday, Ubuntu—a distribution widely used by less advanced users—still listed all its versions as vulnerable. The weak link now lies with those who put off updating and restarting their servers.
The flaw exists in how GNU/Linux handles new internet connections—the most common type of connection in existence. The attack demonstrated by Anthropic exploits a feature called MPTCP, which allows a single connection to utilize multiple networks simultaneously. Programs written in the Go programming language have had this feature enabled by default since version 1.24 (whenever the system supports it). If you manage a server: update your kernel as soon as your system provides the patch, and then restart.
Disabling MPTCP can help in the meantime, but it only reduces the risk—the underlying flaw remains present in standard connections. Anthropic has not published the specific attack method, and there were no reports of it being exploited in the wild as of yesterday, so those who update today will stay ahead of the curve. Just imagine how many other bugs might still be hidden out there. AI helps to accelerate the process—but the real "Jedi" are the programmers themselves, who can rapidly synthesize information, make adjustments, and identify vulnerabilities when using these acceleration tools... hahaha!
