B

Baroness Berger (Lab)

Speaking in the House of Lords on 26 October 2026

Debate

Cyber Security and Resilience (Network and Information Systems) Bill

Contribution

My Lords, I have added my name to Amendment 41 because it identifies and addresses a significant gap in our approach to cyber security: the growing capacity of frontier AI models to interfere with systems on which our essential services depend. This amendment would ensure that there is a corresponding duty on developers to report such incidents and to do so in a timely fashion. This has been put forward as a cross-party amendment because technology is advancing so rapidly and the security of our critical infrastructure is a shared responsibility. It demonstrates that, by working together, we can address what is a very serious emerging risk. If it was not already clear, the urgency of this issue has been brought into sharp focus by recent events, even since we last met in Grand Committee to scrutinise this Bill. The past month, as noble Lords may be aware, has been named “Black October” after advanced AI models accelerated the discovery of critical, long-standing security flaws, most notably in the PHP programming language, triggering global emergency patches and widespread server shutdowns. On 22 October, the Government confirmed that 32 government services had been taken offline as a precaution following the discovery of critical vulnerabilities in PHP, which is a programming language underpinning an estimated 70% of the world’s websites. These vulnerabilities were identified with the assistance of frontier AI, and cyber security companies have subsequently reported millions of attempts to exploit them. To be clear, the Government have said that these precautionary shutdowns did not mean that those websites had been hacked, but this incident demonstrates how AI can help uncover long-standing vulnerabilities at unprecedented speeds and how quickly that discovery can trigger a global security crisis. Cyber security experts are warning us of a new era of AI-enabled hacking. This technology has a dual nature. AI can strengthen our defences, but it can also lower the barriers to finding and exploiting weaknesses in systems on which we all depend. This is not a theoretical concern. In July, the UK AI Security Institute identified an incident during frontier AI testing in which AI agents took sustained, unauthorised actions against real people and organisations. In the most serious case, an agent attempted a software supply chain attack, submitting malicious code and creating fake accounts to pressure a software maintainer into accepting it. Although no real-world harm resulted, the incident demonstrates exactly why we need effective safeguards and clear reporting obligations when AI systems behave in unexpected and potentially harmful ways. We already know the consequences when our digital systems fail. The cyber attack on Synnovis disrupted blood testing and forced the cancellation of hospital appointments and operations. The ransomware attack on the British Library caused extensive disruption to its services, with recovery requiring a prolonged effort. These attacks pre-dated the latest developments in frontier AI, but we must take seriously the possibility that AI will make future attacks faster, more sophisticated and more damaging. The question before us is therefore straightforward. If a frontier AI model gains unauthorised access to or interferes with a regulated organisation’s systems, who ensures that that incident is reported promptly? We know from the example that has been set out that it took 84 days for OpenAI to inform the Australian Government about its model’s breach, and just this weekend Anthropic slipped out an announcement to let us know that Claude has accessed publicly available voter data without any human authorisation. These events are taking place on a daily basis. The organisation affected may not know that AI was involved. The developer, meanwhile, may hold crucial information about the model’s behaviour, the nature of the threat and the steps needed to mitigate it. Without a clear reporting obligation, that information is not reaching the people who need it when they need it. This amendment addresses that gap. It would require an initial notification within 24 hours, a fuller account within 72 hours and a final report within 14 days, and ensure that the affected organisation, its regulator, the relevant cyber security incident response team and the AI Security Institute receive information essential to containing an incident and preventing its recurrence. These are practical and proportionate requirements. They would not prohibit innovation but establish a basic principle of accountability. Those developing powerful technologies must at the very least take responsibility for reporting serious incidents involving their models when an incident has occurred. Without meaningful penalties for non-compliance, these obligations risk being little more than words on paper. During last Thursday’s Cross-Bench debate on AI and children, a point was repeatedly made that is worth repeating here. AI is currently less regulated than milk. The risks to children and the risks to our critical infrastructure and government systems are of course different, but the underlying principle is the same. Our regulatory frameworks must keep pace with increasingly powerful technologies. We are already at the point where harm has become widespread. We need to establish reasonable safeguards, clear responsibilities and effective oversight.

More from Baroness Berger (Lab)

Other recent Hansard contributions by the same speaker.

About Hansard

Hansard is the official verbatim record of proceedings in the UK Parliament. Every word spoken in the Commons and Lords is recorded and published — this page is a single contribution from that record.