UK: Critical Infrastructure Companies Told to Expect an AI Malware Attack by 2028

The UK National Cyber Security Centre (NCSC) has issued its “Annual Review” for the year, and the central theme is hardly anything revolutionary: AI is about to drastically improve attacker capability, with the most skilled attackers seeing the greatest benefit. However, it does have a specific warning within its 24 pages that is worthy of note: the UK government is telling its critical infrastructure firms to expect an AI malware compromise by 2028, most likely conducted by a lesser profit-seeking criminal outfit looking to hold them to ransom.

This prediction reflects both the increasing capability and availability of frontier AI, and the expected arc of proliferation of AI-controlled devices and movement of critical infrastructure functions to cloud services over the next year or so. It is also yet another stark warning to address lingering vulnerabilities as soon as possible, and become comfortable with using AI defenses to counter emerging AI offensive capability.

Critical infrastructure faces sharp and sudden defensive mobilization

One of the central points of the report is seemingly self-evident: the biggest threat posed by AI comes from the best existing hackers, who will use it as a force multiplier. The more interesting detail is the expected timetable of how more common for-profit ransomware and extortion gangs will progress with these tools available. Right now the limit on any criminal immediately going out and making use of AI for attacks and malware is the neutering of publicly available model capability in this area (ie the publicly available version of Mythos vs the one still in private testing), combined with safety guardrails that the world counts on the AI developers to maintain. Underground subscription services that promise jailbroken access to the latest models prove that those guardrails are unsteady at best, however.

NCSC thinks that criminals will be leveraging AI malware by 2028, and that they will manage to hold critical infrastructure to ransom with it. Under UK regulations, the critical infrastructure industry is being required to implement new defensive, monitoring and reporting measures throughout 2027 (and the definition of “critical infrastructure” is even being expanded to encompass new elements such as charging stations and data centers). However, this may not keep pace with both criminal capability and expected proliferation of potential targets.

To that latter point, the NCSC also does note the main threat remains one that has now been in place for years: technical debt and failure to keep up with cybersecurity fundamentals. The report notes that AI is not really creating a variety of new attack vectors, but simply making it faster and easier to discover and exploit the old ones. Timely patching and cybersecurity hygiene continues to go a very long way in addressing this threat.

AI malware will likely still require a human touch

The frontier AI models are doing extremely well in uncovering vulnerabilities. What they are still not doing so well in is crafting complex custom malware for the user. The NCSC prediction does not necessarily indicate that capability will shift dramatically in roughly the next year, but more that there will be a proliferation of softer targets that are more suited to automated hacking. Still, the anticipated attacks on critical infrastructure will likely be from the world of at least fairly experienced and skilled hackers as the attacks continue to require human orchestration.

The greatest threat will thus remain the state-backed APT groups that are more interested in espionage and quiet long-term positioning than noisy disruptive ransomware attacks that carry the possibility of starting a shooting war. But, one way or another, NCSC expects the skill and availability gap for private criminal groups to close in the near term sufficient to be a real threat to critical infrastructure.