The UK's AI Security Strategy is a Masterclass in Trusting the Fox

I have seen this movie before. The script is always the same: a new, disruptive technology arrives, the industry promises to self-regulate in the name of innovation, and the government decides that the safest way to handle the risk is to let the architects of the system mark their own homework.

That is precisely the play currently unfolding in the UK Parliament. As first reported by The Register, the UK government has turned down requests from House of Lords members to include frontier model developers and AI vendors in the scope of the Cyber Security and Resilience (Network and Information Systems) Bill.

Instead of imposing statutory obligations on the companies building the engines of AI, the government is opting for a regime of voluntary safeguards and 'technology-agnostic' requirements. In short: the users are on the hook, but the vendors are off the leash.

### The 'Technology-Agnostic' Shield

During a Tuesday session of the Grand Committee, Cybersecurity Minister Baroness Lloyd of Effra laid out the government's reasoning. According to The Register, Lloyd argued that regulating AI vendors would not actually stop hostile actors from misusing AI products. To the government, the danger isn't the tool itself, but who is holding it.

Consequently, the bill is designed to be technology-agnostic. Rather than regulating the technology providers, the bill focuses on imposing stricter cybersecurity requirements on 'key organizations'—the entities that actually deploy these systems in the real world.

To illustrate this, Lloyd used a hypothetical healthcare organization to show how regulated bodies must secure systems containing AI. This logic led to a pointed exchange with Baroness Kidron, a Crossbench peer and digital rights campaigner. As reported by The Register, Kidron questioned whether this meant the NHS would be tasked with protecting itself while the AI attacking it would have no duties or obligations under the bill to 'check itself' before deployment.

### Voluntary Codes vs. Hard Red Lines

When you remove vendor accountability from the law, you are left with 'voluntary' frameworks. Baroness Lloyd pointed to the voluntary AI Cyber Security Code of Practice and the support of the AI Security Institute (AISI), which tests models before release. She also cited the global AI cybersecurity standard, ETSI EN 304 223, as evidence of the UK's leadership in shaping international technical standards.

But for the peers in the House of Lords, 'voluntary' is just another word for 'optional.' Lawmakers raised several alarms about the current trajectory of AI development, citing reports of rogue agentic behavior involving OpenAI and Anthropic. They also pointed to warnings from Bill Gates, who has expressed concern that commercial incentives are driving AI development forward without sufficient safeguards.

Even the vendors themselves are providing the ammunition for the regulators. Lord Tarassenko, a Crossbench peer and AI researcher, referenced an open letter from OpenAI warning that AI-orchestrated cyberattacks will soon become too prevalent to handle. While some critics dismissed the letter as alarmist rhetoric from companies that profit from the tech, the peers argued that the warning only strengthened the case for actual regulatory intervention.

### The Rejected Kill-Switch

Perhaps the most telling part of the government's stance is its rejection of emergency powers. The House of Lords proposed amendments that would have required AI vendors to prove their products could not cross specific 'red lines,' such as assisting in the development of chemical weapons or evading human oversight.

Furthermore, the government rejected a proposal that would grant the Secretary of State the power to order the shutdown of a widely deployed AI system or a datacenter during a security or operational emergency.

Baroness Lloyd's counter-proposal is far more diluted. Instead of a vendor-level shutdown, the government would direct regulated entities—such as datacenter operators, but notably not the AI vendors themselves—to cease using a specific AI model if it presented a qualifying risk. As Lloyd explained via The Register, the government believes directing a power station to stop using a specific model is a more 'proportionate' response than shutting down complex, distributed datacenter ecosystems.

### The Bottom Line

The Cyber Security and Resilience Bill, first proposed in the 2024 King's Speech and introduced in November 2025, was already courting controversy due to proposed daily fines of £100,000 for organizations that fail to protect against specific threats. Now, it appears the bill will double down on that philosophy: penalize the operator, ignore the manufacturer.

Baroness Kidron summed up the skepticism of the House of Lords perfectly when she asked if we had not already learned from previous experiences in privacy and online safety that allowing tech companies to set their own homework endangers national security.

By treating AI as a neutral tool rather than a product with inherent systemic risks, the UK government is essentially betting that the industry's internal ethics will hold steady even when they clash with the bottom line. In the world of frontier AI, where commercial incentives are moving at light speed, relying on a 'voluntary code' isn't a strategy—it's a prayer.