Can Assurance Help Build AI Systems That We Can Trust?
The Partnership on AI co-hosted the AI Standards Hub Global Summit in Glasgow with partners including The Alan Turing Institute and the British Standards Institution. The event gathered experts to discuss building assurance infrastructure for safe, reliable, and accountable AI systems. Four key themes emerged from workshops and discussions. First, assurance must extend beyond pre-deployment to include real-time post-deployment monitoring, which is currently underutilized. Second, demand for independent external assurance remains low due to unclear regulations and proprietary concerns, though participants favored legislation and transparency measures like incident reporting to boost engagement. Third, frontier AI models require state-of-the-art, adaptable evaluation standards to address high-stakes risks such as CBRN threats, emphasizing process standards over rigid prescriptions. Finally, the summit highlighted the need for trusted independent assessors with the authority to communicate expert judgments. The discussions aim to strengthen the AI assurance ecosystem, enabling citizens and enterprises to adopt AI with calibrated trust by understanding both its capabilities and limitations.
Wire timeline
Can Assurance Help Build AI Systems That We Can Trust?
The Partnership on AI co-hosted the AI Standards Hub Global Summit in Glasgow with partners including The Alan Turing Institute and the British Standards Institution. The event gathered experts to discuss building assurance infrastructure for safe, reliable, and accountable AI systems. Four key themes emerged from workshops and discussions. First, assurance must extend beyond pre-deployment to include real-time post-deployment monitoring, which is currently underutilized. Second, demand for independent external assurance remains low due to unclear regulations and proprietary concerns, though participants favored legislation and transparency measures like incident reporting to boost engagement. Third, frontier AI models require state-of-the-art, adaptable evaluation standards to address high-stakes risks such as CBRN threats, emphasizing process standards over rigid prescriptions. Finally, the summit highlighted the need for trusted independent assessors with the authority to communicate expert judgments. The discussions aim to strengthen the AI assurance ecosystem, enabling citizens and enterprises to adopt AI with calibrated trust by understanding both its capabilities and limitations.
Partnership on AI