True Innovation Requires Responsibility: Our Three Priorities for 2025

Over the last few months, we have seen a flurry of new AI models and capabilities. For those of us who believe that true innovation requires responsibility, these releases are reminders of the importance of working together so that AI advances positive outcomes for people and society. At PAI, we are fortunate to bring together a remarkable multistakeholder community to explore new questions and develop novel methods to advance understanding, better practice and inform public policy. This year, our three priorities will focus on understanding the evolving AI landscape with agents, supporting vibrant assurance and accountability ecosystems, and reaching out to new partners and regions globally.

Agentic AI Risk Management

This year we expect to see a significant shift towards the development and deployment of AI agents. Agentic AI refers to AI systems that act on behalf of users, acting with some degree of autonomy to achieve goals without human intervention or guidance. The goals of agents are to understand a user’s general goals and utilize context to solve specific problems without explicit instructions. Examples range from providing customer service, where AI can go beyond a chat bot and execute a service for a customer to making a dinner reservation on behalf of a user through a third party app and scheduling it on their calendar.

Building directly on our work on safe foundational model deployment, with seed funding from CSET at Georgetown University, our unique community of technologists, researchers, civil society, and academic experts will work on monitoring impacts of AI assistants and identifying the landscape of potential risks and harms of agentic AI. Understanding the emerging impacts of agents on our guidance in media integrity, labor and the economy and public policy will be central to our 2025 plan.

AI Accountability Ecosystem

In 2025, we will focus on defining the elements of a robust AI assurance ecosystem, and bridging important but fragmented efforts across policy, advocacy, research, and industry. To unlock innovation across sectors, there is a need for shared norms and standards in foundational elements from documentation and corporate disclosures to risk and impact assessments, standardized metrics, technical standards, certifications, and evaluations. Central to this work is the participation of citizens and communities in designing AI and the terms of its use.

PAI will host a side-event at the AI Action Summit in Paris to catalyze this work. Experts from across our Partner community will lay out the core components of an effective ecosystem and provide a platform for stakeholders to collaborate on the practical steps needed to build it, particularly as we work towards advancing open AI models. Fostering an ecosystem of transparency requires the creativity of stakeholders from across sectors, including investors, policy-makers, and workers, who play a vital role in shaping the future of AI. In doing so, we can ensure that AI is developed in a way that is accountable to citizens and accessible to all.

Global Connectivity

Global AI governance requires a truly international community. This year, we will build bridges with our Partners across regions, sectors, and perspectives. With a focus on expanding efforts outside of North America and Western Europe, we are dedicated to seeing AI developed in the interests of a diverse set of cultural and geographic contexts. We will launch a new international expert advisory council, SAIGE, to provide sustained guidance on emerging topics, the current state of AI safety, and PAI’s agenda-setting and impact.

As we meet with policymakers from across the globe, we emphasize the importance of interoperability in effective policy development. By participating in global fora such as the AI Action Summit, we will underscore the need for alignment on best practices, especially those that have been developed via a multistakeholder process.

What’s Next?

This week, I will join several PAI colleagues and partners at the AI Action Summit in France. We will host a pre-Summit event providing a unique opportunity to bridge fragmented efforts across policy, advocacy, research, and industry. With a focus on enabling the development of open, accountable, and fair AI systems, we are bringing together diverse stakeholders to engage in dynamic discussions and collaborative sessions that underscore transparency, sustainability, and responsible innovation.

Partnership on AI views the AI Action Summit as a unique and timely opportunity to accelerate the international dialogue on AI developed in the public interest. We are calling on national leaders to ensure that AI is adopted in the public interest globally. The Summit is a key moment for governments to align on governance initiatives so that AI is adopted to benefit people and society. In the Leaders’ Declaration, we hope to see an emphasis in 4 key areas:

• multi-stakeholder action and citizen participation;
• investment in open research, innovation and accountability measures;
• support for workers and skills development across the AI value chain;
• And a clear plan for global governance that drives forward institutional alignment, accountability and interoperability across countries including through an assurance ecosystem