The 30-Minute Exploit: What AI-Assisted Attackers Mean for OT Remote Access

Key takeaway: A joint NSA, CISA, FBI, DOE, and EPA advisory confirms threat actors are using AI-generated exploitation scripts against internet-exposed Siemens PLCs, disguised as routine monitoring software and using valid vendor credentials. The real shift isn't sophistication, it's speed: AI has compressed reconnaissance and exploitation from weeks to minutes, so a passed authentication check no longer confirms a session is trustworthy. OT remote access has to close that gap by scoring trust continuously through the whole session, not just at login, and by removing the static infrastructure that reconnaissance depends on. 

Authentication used to be the finish line. AI just made it the starting gun. Here’s what that means for how OT remote access has to work now. 

A prospect asked me a version of this question a few weeks ago, and I’ve been asked it enough times now that I think it’s the right question for this moment: “If someone got in tomorrow using a real vendor’s credentials, how long would it take them to find something worth touching?” 

Two years ago, that was a multi-week question. It required a skilled human, patience, and a fair amount of trial and error against an unfamiliar environment. Today, the honest answer for a lot of OT environments is: not long. Minutes, not weeks. And the gap between those two numbers is the entire subject of this post. 

The Federal Government Just Confirmed It 

On August 19, the NSA, CISA, the FBI, the Department of Energy, and the EPA issued a joint advisory confirming that threat actors are using AI-generated exploitation scripts against Siemens S7 Series programmable logic controllers across U.S. critical infrastructure. The scripts are disguised as legitimate OT monitoring software, built using internet scanning tools to find exposed devices, and aimed squarely at Critical Manufacturing, Energy, Water and Wastewater, Chemical, and Food and Agriculture. The agencies were direct about why it matters beyond Siemens specifically: “Ongoing PLC targeting activity is broader than Siemens PLCs. All PLC owners and operators should apply relevant mitigations.” 

This isn’t a hypothetical about some future model. It’s a description of what happened this month, following what happened in July when more than 30 municipal water systems were hit across a single weekend in Minnesota, and what occurred earlier in the spring when Iranian-linked actors targeted internet-exposed Rockwell Automation controllers. The pattern connecting all three: no exotic malware, no novel zero-day. Exposed devices, weak authentication, and now, AI doing the reconnaissance work a skilled human used to have to do by hand. 

It's Not Sophistication. It’s Time. 

The World Economic Forum published a piece in July, co-authored by Siemens Energy’s global head of industrial cyber and digital security, that put a specific number on this shift: AI agents have already been shown to discover and exploit vulnerabilities in as little as 15 minutes, citing reporting on the first fully LLM-driven ransomware campaign. The same piece cites Siemens Energy’s own survey finding that only 46% of security practitioners believe their organization’s OT environment is adequately protected, and that typical OT defenders take more than a month to detect a breach and over seven months to recover from one. 

Sit with that gap for a second. Fifteen minutes to find and exploit a vulnerability against more than a month to notice it happened. Fifteen minutes isn’t the OT-specific number yet; that research describes software exploitation broadly, and industrial environments have their own friction that can slow an attacker down. But the direction is not in question, and thirty minutes is a reasonable, conservative translation of that same curve into an OT context: still fast enough that whatever advantage your team had from an attacker needing time and skill is now gone. 

Andrew Costis, who leads adversary research at AttackIQ, made the sharper point in his read on the Siemens advisory: “This latest exposure poses a bit of a different threat. The danger is more immediate because PLCs sit much closer to the physical processes that keep critical infrastructure running.” His conclusion tracks with what we’ve been telling customers all year: “Patch management and segmentation are important first steps, but they can’t be the finish line.” 

Authentication Answers the Wrong Question Now 

Here’s the uncomfortable part. Every one of these advisories describes attackers using credentials that check out. Valid logins. Legitimate-looking software. The front door isn’t being broken down, it’s being walked through, because the AI doing the reconnaissance and the credential handling is good enough now to look like routine vendor activity. 

That breaks a quiet assumption a lot of OT remote access programs have been running on: that if MFA passes and the credential is valid, the session is trustworthy. It was a reasonable assumption when faking that combination took real skill and time. It stops being reasonable the moment an AI agent can map your environment, mimic a vendor’s normal behavior, and chain an exploit together in the time it takes to get coffee. 

We covered the front half of this shift back in June, in our “OT Secure Remote Access in a Mythos World” blog: AI that finds and exploits vulnerabilities faster than skilled humans, and the session-visibility gap most programs still have. What’s changed since then isn’t the thesis, it’s the evidence. We’re not talking about a lab benchmark anymore. We’re talking about a federal advisory covering live activity against the systems this audience runs. 

Two Things Have to Move at the Pace This Threat Requires 

If credentials can be faked convincingly and reconnaissance now happens in minutes, two things have to change about how OT remote access works. 

• Trust has to be evaluated for the whole session, not just at the door. A valid login says a credential checked out. It doesn’t say the behavior after that login matches the vendor who’s supposed to be on the other end of it. Dispel Intelligence scores every session continuously, login through disconnect, against behavioral baselines built only from trusted activity, so a compromised account can’t train the system to treat anomalous behavior as normal. A green login on a red day still gets flagged. 

• The thing being mapped has to stop sitting still. Frontier AI has made exploiting a static network pathway a 30-minute operation. Moving Target Defense in the Dispel Zero Trust Engine eliminates the stable attack surface that operation depends on. There’s no persistent IP or port to map, because every connection runs on disposable infrastructure: a single-use pathway, encrypted end to end, micro-segmented, MFA-verified, destroyed the moment the session disconnects. No dwell time, no attack. An AI agent can scan for thirty minutes and find nothing worth returning to, because what it mapped yesterday won’t exist tomorrow. 

There’s a third piece worth naming given what these advisories describe: attackers disguising themselves as legitimate vendors using valid credentials. Dispel Identity closes the part of that gap neither of the above two fully touches. Confirming a real, verified person, government ID and biometric match to the NIST IAL2 standard, is actually the one behind the account. Credential mimicry works because a password and an MFA prompt can’t tell you who’s typing. Identity verification can. 

This Isn’t a Call to Panic. It’s A Call to Move the Finish Line. 

None of this is an argument that OT is uniquely doomed, or that the progress the industry has made on MFA, segmentation, and vendor access restrictions was wasted. It wasn’t. It's the necessary foundation. It's just no longer the finish line, because the thing it was built to stop, a slow, skilled human attacker, isn’t the primary adversary anymore. 

The constructive version of this story is straightforward: the same automation that's compressing attacker timelines can compress your response timeline too. Continuous session scoring instead of a login check. Infrastructure that moves instead of infrastructure that waits to be found. That's not a bigger version of what OT remote access already does. It's the same speed the threat is now operating at, applied to the defense instead. 

That's the sanctioned path made real. When trust is verified continuously and the infrastructure never sits still long enough to map, the secure way to connect and the fast way to connect are the same path. Vendors aren't slowed down proving who they are. Attackers impersonating them don't find a static target waiting to be scanned. Nobody has to choose between speed and security, because the architecture that removes friction for your team is the same one that removes the head start for anyone trying to be your team. 

A Critical Control for Modern Risk 

Secure remote access alone delivers more than 12% risk reduction, according to the Dragos 2025 OT Cybersecurity Financial Risk Report — and it's exactly what SANS ICS Control 4 calls out as a critical safeguard for OT cybersecurity. Download the SANS report →