IndustrialCyber

Inside OT’s Fundamentals Gap: A PLC Fault, an AI Exploit, and Leaked Identities


On October 27, Google published its final incident report on the London power failure that took its europe-west2-b zone offline a week earlier, disrupting Revolut, Lloyds Banking Group, HSBC, and the video links used for remote court hearings in England and Wales. The cause was not an attacker. According to Google, the undervoltage sensing inputs on the programmable logic controllers that start the facility’s generator plant were wired on the line side of a utility breaker that had been isolated for repairs. When the live utility feed failed, the controllers still saw voltage, and the generators never received a signal to start.

On a recent episode of my CyberSecureOT podcast, I called this outage a wake-up call: we spend so much time worrying about zero-day exploits and malware that we forget the fundamentals, and hardware needs power. Google’s final report shows that the fundamental that failed was an industrial control system. The same theme runs through the episode’s other stories: an AI-written Linux root exploit and identity leaks at Capcom and the U.S. Department of Homeland Security. In each case, OT security fundamentals decided the outcome: how control logic is tested, how quickly exposed systems are patched, and how identity data is protected.

A generator PLC, not an attacker, darkened a London cloud zone

Google’s report describes a chain of ordinary conditions. Utility feed B had been opened and locked out for repairs after an October 14 generator load test, leaving the facility on feed A alone. When an underground cable fault on the utility network outside the facility took out feed A, both the primary and backup PLCs read live voltage upstream of the isolated feed B breaker and concluded a healthy source existed. The static UPS batteries, which Google says provide five minutes of runtime, drained, and hosts, network switching, and storage in the zone lost power.

Utility power came back within the hour, but Google puts customer impact at six hours and eight minutes. Google says sudden demand on the region’s other two zones, combined with cross-zone control-plane calls that timed out, left the wider region degraded, with close to 40 services affected.

Google’s remediation list deals almost entirely with control system discipline: PLC code reviews with vendors before generator testing, validation of PLC configurations at sites with similar topologies and PLC manufacturers, a maintenance freeze, and, in a November 4 revision, a commitment to validate PLC code during facility commissioning by December 18.

On the podcast, I argued that true resilience requires geographic redundancy, spreading critical infrastructure across multiple isolated power grids, and that when a bank cannot stay online through a local power failure, the public has reason to ask how it would handle a targeted attack. Google’s account of cross-zone dependencies bears out the first point: losing one zone degraded its neighbors. Operators running historians or remote monitoring in a single cloud region carry the same dependency, which an earlier article in this series on IT-OT dependency examined in a manufacturing context.

What an AI-built Linux root exploit means for OT patch windows

On October 27, Anthropic’s Frontier Red Team reported that an unreleased internal research model, after initial prompting, wrote an exploit for CVE-2026-98432, a race condition that leads to a use-after-free in the Linux kernel’s networking stack. Against a server with a Multipath TCP (MPTCP) listener, the exploit leaked a kernel pointer to defeat kernel address space layout randomization in 48 minutes, then reached root command execution about five minutes later, with no credentials and no user interaction. Anthropic did not publish the exploit code, the kernel version, or the test configuration.

On air, I said this was not a human hacker spending months writing custom code, and that is the part that matters. Anthropic’s own figures put the work at roughly 53 minutes after initial prompting, not seconds, but that is still a short path from a disclosed kernel bug to unauthenticated remote root.

The patch picture is uneven. The Linux kernel CVE team scored the flaw 9.8, while The Hacker News reported Red Hat, SUSE, and Oracle rating it lower. Debian shipped fixes on October 16 and 17. Ubuntu’s CVE page, updated November 14, shows fixed kernels for 22.04, 24.04, and 26.04 LTS, with 20.04 still vulnerable. As of late October, no exploitation in the wild had been reported, and the flaw was not in CISA’s Known Exploited Vulnerabilities catalog.

On industrial networks, Linux runs inside routers, gateways, and protocol converters whose kernels are set by the device vendor. Earlier Forescout and Finite State research found OT and IoT router firmware averaging 20 exploitable n-day kernel vulnerabilities. The speed problem is already reaching controllers too: Claroty’s Team82 used an AI coding agent to rebuild Danfoss controller exploits in 28 minutes. As I said on the episode, the era of AI-driven exploitation is here, and our defenses need to evolve just as fast. Anthropic’s exploit required a reachable MPTCP listener, which makes exposure, not only patch status, the first question to ask of any Linux-based device on an industrial network.

Identity leaks are the raw material for the next intrusion

Capcom disclosed on October 22 that a third party had accessed CAPCOM ID, its membership service, exposing 11,774,247 records containing email addresses and account metadata: creation dates, last-login times, login counts, and the IP address used at last login. Passwords were changed without authorization on 196 accounts, which Capcom reset. Capcom has not disclosed how the attacker got in. On the podcast, I said breaches like this usually come down to a misconfigured database or a compromised internal credential; for Capcom, that remains my assessment rather than a finding. The larger risk is what follows: email lists feed phishing and credential stuffing against people who reuse passwords, so a reused password needs changing everywhere it was used, not only at Capcom.

The DHS leak raised the stakes. On October 28, Distributed Denial of Secrets released data on more than 41,000 DHS employees and contractors, attributed to a hacktivist group calling itself the Department of Peace. The records include names, email addresses, phone numbers, job titles, and office locations, and the BBC counted more than 21,000 Customs and Border Protection staff. DDoSecrets co-founder Emma Best said the data was harvested through the Okta single sign-on service DHS uses. Reuters could not corroborate that claim, and Okta told WIRED there was no evidence of a compromise. A DHS spokesperson said the department was working “to address the situation swiftly and effectively.” On air, I suggested the route was likely a weaker outside vendor, which is how exposures like this often happen; how the data was actually obtained has not been established.

The industrial connection is direct. The FBI and Secret Service recently urged critical infrastructure operators to secure Fortinet gateways after attackers used credential stuffing and password spraying, drawing on earlier leak dumps and infostealer logs, to harvest access across all 16 sectors. Staff directories and password-reuse lists are the inputs to that kind of campaign. As I put it on the show, your data is only as secure as the weakest link holding onto it.

Where OT security fundamentals need attention next

The Google report sets dates worth tracking: validation of PLC configurations at similar sites by November 15, generator load transfer tests on restored feeds by November 17, and the commissioning commitment by December 18. Operators who rely on colocation or cloud facilities, or who run their own substations and backup generation, can ask a direct question: has the generator start logic been tested with one utility feed locked out?

On Linux, the immediate work is inventory and exposure: which industrial assets run Linux, which expose listeners such as MPTCP, and when each vendor will ship a fixed kernel. On identity, the Capcom and DHS data will keep circulating, so multi-factor authentication on every remote access path into OT is what turns a leaked password into a failed login rather than an intrusion.

To hear the full discussion, listen to the episode on my CyberSecureOT podcast.



Source link