Jerry: Factory Admin set his password to an insult aimed at hackers. A hacker did eventually read it. This is the daily DefSec brief for Friday, October 30th, 2026. If you do only one thing today, upgrade your self-managed GitLab servers to the latest patch. One, attackers are now exploiting a GitLab flaw we covered earlier this month. Any account that can push to a public project can run code on a self-managed GitLab server. With open signups turned on, that's anyone on the internet. Alpine Linux says a newly registered account went after its GitLab server on October 21st. Alpine rebuilt the server and reset every password and GitLab's secret keys. The fix has been out since October 12th. Upgrade to the latest patch release and turn off open sign-up if you don't need it. If something from this week's show changed a plan at work, Tell me at feedback at defensive security dot org. Two, no attacks yet, but a Linux kernel flaw gives a remote attacker a root shell on servers With a multipath TCP listener. Anthropic's model wrote a working exploit. Anthropic hasn't released it. Programs built with recent versions of Go open those listeners by default. Ubuntu and Red Hat haven't shipped a fix yet. The kernel team's own severity score assumes the attacker sends IP options with the connection. Normal traffic almost never does. I think dropping inbound packets with IP options at your firewall is a reasonable stopgap. Turning off multipath TCP with a kernel setting also closes the path anthropics exploit used. Neither is confirmed against every variant. Watch your distribution for better guidance and we'll bring you an update when there is one. Install the fixed kernel as soon as it ships. Three, an update on a WordPress flaw from earlier this week. WordPress fixed a flaw that lets anyone take over a site during the few seconds a core update takes. In the first day, with more than 18,000 requests, WordFence now sees attackers trying it. No compromised sites are confirmed yet. Patch stack has called it impractical to exploit. Attackers are trying anyway. Update WordPress core and block web access to the upgrade folder. Four. Exploit code is public for the browsers built into five AI coding tools. In Cursor, a web page got the agent to run a command with one keypress. In Codecs, VS Code, and Claw Desktop, one click on a link in a rigged project could expose any file on the disk. Codecs and VS Code are still unpatched. Microsoft points to its workspace trust prompt. The researcher says everyone just clicks trust. Keep long live secrets off of developer laptops. Five, attackers are using a new Windows implant that sends what it steals to OpenAI for analysis. If the operator brings no key, it uses one it finds on the infected machine. The victim pays for the attacker's AI and OpenAI's logs show the victim's key. Alert on open AI API calls for machines with no reason to make them. Six. Wired reports that an Accenture employee gave a personal clawed code agent their own login to a defense department network. For about two months on that network, the agent wrote and deployed code. It appears to have sent passwords and a network map to an outside website. Accenture says an investigation found the allegations without merit. Give AI agents their own accounts. with only the access they need. 7. While posing as AI crawlers from OpenAI, Anthropic, and others, attackers are scanning web servers. One actor cycled through more than 30 fake bot names asking for.emv files, cloud credentials, and SSH keys. Sites that let AI crawlers through by name give it the same pass. Block web requests for.env files and key files. Eight. No attacks yet, but researchers disclosed six ML flow flaws, including one that lets anyone who reaches the server send requests from inside it. Those requests can reach the cloud metadata service. ML flow ships with login turned off. On a default install, anyone who reaches the server can use the three permission flaws too. Upgrade ML flow and turn its login on. Nine, no attacks yet, but Chrome is another security update. It's third in nine days. With one critical fix. Pushed the update and forced the restart. ten. A factory hired a pen tester after a breach. He found one Windows 7 server on a milling machine still on the domain and still open to Eternal Blue. He pulled the IT manager's admin password from the server's memory in plain text. It was hey hacker, go F yourself in LeetSpeak. He sat on it for a week, sure it was a trap. It was real, and it gave him the whole network. Keep admin passwords off of legacy systems. The message did reach a hacker, just one with a contract. Once more, upgrade self-managed GitLab and keep an eye on that Linux vulnerability. Thanks for watching. Sources are in the show notes. Like, subscribe and share the show with a friend who'd use it. I'm Jerry Bell and I'll be back again on Monday.