Thank you, Peter, and my gratitude extends to the entire team at Imperial College for graciously hosting us today.
I’m equally appreciative of all of you for attending this important gathering.
Reflecting on the Industrial Revolution
Two centuries ago, Britain found itself entrenched in the throes of the Industrial Revolution. This transformative period saw the emergence of vast new industries, the accumulation of immense fortunes, and the introduction of groundbreaking technologies that significantly altered daily life.
However, the benefits derived from this revolution were far from equitably distributed. In mills and factories throughout the nation, children were subjected to appalling working conditions. Although Parliament enacted some protective legislation, these laws were largely ineffective due to the government’s inability to enforce them.
Consequently, in 1833, the factory inspectorate was established, initially comprising just four inspectors tasked with overseeing thousands of mills. This modest inception was inadequate to address the enormity of the challenge, yet it laid the groundwork for a crucial principle: when technological advancements shift the balance of power in society, it is the government’s responsibility to safeguard those who bear the brunt of such changes.
Embracing Technological Change for Humanity
Importantly, fulfilling this responsibility requires more than simply enacting new laws. It necessitates the creation of institutions equipped with the expertise to take action. Progress must benefit the individuals whose work underpins it, and technological advancements should enhance human dignity.
We currently stand on the brink of yet another technological revolution, one whose repercussions will resonate for generations. As we transition into the era of superintelligence, the decisions we make now hold significant weight. I am committed to a straightforward principle: we must develop machines that serve humanity.
These machines should facilitate scientific discoveries in medicine, grant care workers more time with those who rely on them, and enhance the productivity of British businesses. Most importantly, they must ensure that humans remain in command.
Government Commitment to AI Safety
However, this vision requires swift action. The British public rightly anticipates that the government will take the potential risks associated with AI seriously, as our foremost duty is to ensure the safety of our nation.
This understanding is why the Prime Minister has granted me a position at the cabinet table, signalling that Britain is taking artificial intelligence seriously. While we acknowledge its immense potential, we are equally resolved to address its associated risks.
I empathise with the natural anxieties surrounding job security and the implications of AI for future generations. I share these concerns. While I cannot promise that we can halt change, I assure the British public that we will not abandon working individuals to shoulder the burdens of this transformation. I will collaborate tirelessly with my colleagues to secure a brighter future for our children and grandchildren.
Proactive Measures Against Emerging Risks
In recent weeks, the British public has been exposed to extraordinary forecasts from both corporations and employees at the forefront of AI technology. It is entirely reasonable for the public to expect that the government will heed these warnings with utmost seriousness, and we certainly do.
Our responsibility begins with protecting citizens from emerging social harms. I take pride in having been the minister who prohibited the abhorrent use of AI to create intimate images of women and girls without their consent. Alongside the Secretary of State, I stood firm against the wealthiest individuals in the world, successfully upholding our nation’s position. Furthermore, I was proud to announce a social media ban for individuals under 16, prioritising our children’s futures over corporate profits.
Whenever new harms manifest, we have acted decisively, and we will continue to do so. I am working closely with my colleague, Secretary of State Lisa Nandy, to ensure our proactive approach remains effective.
Global Cooperation in AI Regulation
It is crucial to acknowledge that many challenges presented by new technologies extend beyond our borders. With increasingly capable models being openly distributed, isolation is no longer an option.
This recognition is why the Prime Minister has made artificial intelligence a focal point of Britain’s G20 Presidency next year. We understand that these challenges cannot be tackled by a single nation alone, but Britain, with its capabilities and trust, will play its essential role.
The AI Security Institute will significantly contribute to shaping our collective understanding of technical standards. In fact, shortly after this address, we will be publishing an initial perspective on the areas where we seek progress.
Investing in Domestic AI Capabilities
As we engage in global cooperation, we must also take decisive action at home, cultivating a distinctly British approach that embodies the pragmatic and forward-thinking nature of our nation. Throughout this process, my unwavering priority will be the safety of the British public.
We will employ every available power and legislate where necessary. However, we will not rely solely on legislation. British ingenuity, scientific expertise, and technical capabilities will be essential in comprehending these increasingly powerful systems, defending against new threats, and responding effectively when issues arise.
Understanding the nature of the challenges we face is the first step. Britain has a storied history of uniting exceptional minds to confront significant technological challenges, from the codebreakers at Bletchley Park to the scientists and engineers currently working at the AI Security Institute.
Building a Robust Cybersecurity Framework
The AI Security Institute is an outstanding asset to the British state, uniting over a hundred exceptional technical experts committed to understanding these systems and their behaviours. Many of these individuals could command higher salaries elsewhere, yet they have chosen to dedicate their talents to safeguarding our nation, and we should take immense pride in their contributions.
Recently, we have collaborated closely with the US government and leading AI developers, from His Majesty’s gathering at Dumfries House in Scotland to discussions at the United Nations General Assembly in New York, to ensure that the AI Security Institute retains access to the most powerful AI models prior to their release. I can confirm that scientists at the AI Security Institute are currently conducting thorough evaluations of the latest generation of models, including Gemini 4 Argon, to identify potential risks early and to test robust safeguards for public safety.
Preparing for Future AI Challenges
It is vital that our understanding of these systems not only reflects on past developments but also anticipates future challenges. This involves discovering new methods for verifying and discerning capabilities, with hardware-enabled verification drawing on Britain’s strengths in semiconductor design as one promising avenue of research.
By combining our existing expertise in model testing with prospective capabilities in verification, Britain’s unparalleled understanding will serve as the cornerstone of our AI security framework.
While comprehending risks is essential, we must also fortify Britain against potential attacks and enhance our preparedness for instances where AI systems behave unexpectedly. We have seen troubling incidents, such as a recent one in Australia where an OpenAI agent inadvertently accessed restricted information on public spending.
Strengthening Cybersecurity Against AI Threats
Although the Australian government reported no personal medical records were compromised, the incident raises legitimate concerns. It is likely, if not inevitable, that our own cyber defences will face similar breaches. The National Cyber Security Centre has warned that AI will likely enhance the efficacy of cyber intrusions.
In response, my colleague, Security Minister Dan Jarvis, will lead a cross-government initiative aimed at strengthening our defences against cyber threats, particularly those posed by AI. Earlier this year, we published the Government Cyber Action Plan to address these concerns, and we are set to launch the National Cyber Action Plan soon.
If AI equips potential attackers with formidable new capabilities, we must harness the same technology to bolster our defences. Today, I am pleased to share findings from ongoing efforts in this area.
Proactive Measures to Identify Vulnerabilities
Through two government pilots, frontier AI has successfully identified and addressed more than a thousand vulnerabilities alongside other threats to our public services. Our central vulnerability-monitoring service has significantly reduced the time required to rectify identified weaknesses from 50 days to a mere eight.
Building on these advancements, I am announcing that the Cyber Minister, the Security Minister, and I will spearhead the urgent deployment of frontier AI capabilities to detect weaknesses within government systems before they can be exploited, thereby protecting essential services and enabling more effective responses to emerging threats.
My ambition is for Britain to become one of the most resilient nations against attacks using artificial intelligence. Whenever stronger defences necessitate legislative support, we will pursue it. The Cyber Security and Resilience Bill provides us with an opportunity to achieve precisely that.
Collaborative Approach to AI Incident Response
We will work diligently with Parliament and technical experts to determine how we can fortify our framework to better shield Britain from AI-driven cyber threats and loss-of-control risks. This includes identifying and reporting serious incidents and near misses, while ensuring that significant AI incidents can be reviewed appropriately.
Beyond mere defence, we must also prepare for incident response. As AI systems gain wider adoption with reduced human oversight, incidents like the one in Australia are likely to become more commonplace. When such events occur, it is imperative for the government to swiftly ascertain what an AI agent was instructed to do, what information it accessed, its rationale for actions taken, and whether similar failures could arise elsewhere.