[ad_1]
How summits in Seoul, France and beyond can boost international collaboration on frontline AI security
Last year, the UK government hosted the first major global summit on AI border security at Bletchley Park. It drew the world's attention to rapid progress at the frontier of AI development and provided concrete international actions to respond to potential future risks, including the Bletchley Declaration; new AI security institutes; and the International Scientific Report on Advanced AI Safety.
Six months after Bletchley, the international community has an opportunity to build on this momentum and spur further global collaboration at this week's AI Seoul Summit. Below, we share some thoughts on how the summit – and future summits – can drive progress towards a common, global approach to border AI security.
AI capabilities have continued to advance rapidly
Since Bletchley, there has been strong innovation and progress across the field, including at Google DeepMind. AI continues to drive breakthroughs in critical scientific areas, with our new AlphaFold 3 model predicting the structure and interactions of all life molecules with unprecedented accuracy. This work will help transform our understanding of the biological world and accelerate drug discovery. At the same time, our Gemini family of models has already made products used by billions of people around the world more useful and accessible. We've also been working to improve the way our models perceive, reason, and interact, and recently shared our progress building the future of AI assistants with Project Astra.
This advance in AI capabilities promises to improve the lives of many people, but also raises new questions that need to be addressed together across a number of important security areas. Google DeepMind is working to identify and address these challenges through groundbreaking security research. In the last few months alone, we have shared our evolving approach to developing a holistic set of security and responsibility assessments for our advanced models, including early research assessing critical capabilities such as deception, cybersecurity, self-proliferation and self-proliferation. Argumentation. We also published extensive research on aligning future advanced AI assistants with human values and interests. Beyond LLMs, we recently shared our approach to biosecurity for AlphaFold 3.
This work is based on our belief that we need to innovate in security and governance just as quickly as we innovate in capabilities – and that both things must happen simultaneously, continually informing and reinforcing each other.
Building international consensus on disruptive AI risks
To maximize the benefits of advanced AI systems, international consensus must be reached on critical border security issues, including anticipating and preparing for new risks beyond those posed by current models. However, given the high level of uncertainty about these potential future risks, there is a clear need from policymakers for an independent, science-based perspective.
That's why the release of the new preliminary international scientific report on the safety of advanced AI is an important part of the AI Seoul Summit – and we look forward to presenting evidence from our research later this year. Over time, this type of effort could become a central contribution to the summit process, and if successful, we believe it should be given a more permanent status, broadly modeled on the role of the Intergovernmental Panel on Climate Change. This would be an important contribution to the evidence base that policymakers around the world need to inform international action.
We believe that these AI Summits can provide a regular forum dedicated to building international consensus and a common, coordinated approach to governance. Maintaining a specific focus on border security will also ensure that these meetings complement, rather than duplicate, other international governance efforts.
Establish evaluation best practices and a coherent governance framework
Assessments are a critical component required for AI governance decision-making. They allow us to measure the capabilities, behavior and impact of an AI system and are an important input for risk assessments and the development of appropriate mitigations. However, the science of AI border security assessments is still at an early stage of development.
That's why the Frontier Model Forum (FMF), which Google launched along with other leading AI labs, is discussing best practices for evaluating frontier models with AI safety institutes in the US and UK, as well as other stakeholders. The AI Summits could help expand this work internationally and help avoid a patchwork of national testing and governance regimes that overlap or conflict with one another. It is critical that we avoid fragmentation that could inadvertently impact security or innovation.
The US and UK AI security institutes have already agreed to develop a common approach to security testing, an important first step towards greater coordination. We believe there is an opportunity to build on this over time and develop a common, global approach. A first priority of the Seoul Summit could be to agree a roadmap for a wide range of stakeholders to collaborate on developing and standardizing benchmarks and approaches for AI evaluation.
It will also be important to develop common risk management frameworks. To contribute to these discussions, we recently launched the first version of our Frontier Safety Framework, a set of protocols to proactively identify future AI capabilities that could cause serious harm and establish mechanisms to detect and mitigate them. We expect the Framework to evolve significantly as we learn from its implementation, deepen our understanding of AI risks and assessments, and collaborate with industry, academia and government. We hope that sharing our approaches over time will facilitate collaboration with others to agree on standards and best practices for assessing the security of future generations of AI models.
Towards a global approach to breakthrough AI security
Many of the potential risks that could arise from advances at the forefront of AI are global in nature. As we head toward the AI Summit in Seoul and look toward future summits in France and beyond, we welcome the opportunity to advance global collaboration on AI security at the forefront. We hope that these summits will provide a dedicated forum for progress towards a common, global approach. Getting this right is a crucial step in unlocking the enormous benefits of AI for society.
[ad_2]
Source link