Can Assurance Help Build AI Systems That We Can Trust?

2026-03-26 · Source: Partnership on AI · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Emerging Technologies & Innovation · Depth: Intermediate, short

Summary

The AI Standards Hub Global Summit in Glasgow, co-hosted by Partnership on AI and The Alan Turing Institute, British Standards Institution, and UK's National Physical Laboratory (NPL), convened experts to discuss building robust AI assurance infrastructure. Key themes emerged, including the necessity for assurance to extend beyond initial deployment to include continuous post-deployment monitoring, a service currently underutilized. The summit also highlighted low demand for independent assurance services, attributed to unclear regulatory expectations and limited awareness, with 46% of 76 respondents favoring legislation to increase demand. Furthermore, the discussion emphasized that frontier AI models, with their high-stakes risks like CBRN, require state-of-the-art, adaptable evaluation standards, such as process standards. Finally, the rapid adoption of agentic AI systems is outpacing formal standardization, underscoring the need for voluntary frameworks and accelerated development of assurance infrastructure.

Key takeaway

For CTOs and VPs of Engineering deploying AI systems, prioritizing continuous post-deployment monitoring and engaging independent assurers is crucial for mitigating risks and building calibrated trust. Your teams should advocate for and implement adaptable process standards, especially for frontier models, and leverage voluntary frameworks to bridge gaps where formal ISO standards lag, ensuring systems are safe and accountable before widespread public adoption.

Key insights

Comprehensive AI assurance infrastructure, including post-deployment monitoring and adaptable standards, is critical for safe and trustworthy AI.

Principles

• Assurance must extend beyond deployment.
• Standards must be adaptable for frontier AI.
• Voluntary frameworks fill standardization gaps.

Method

The summit identified policy levers to increase demand for external AI assurance, including legislation (46% support) and greater transparency via use case registers and incident reporting (41% support).

In practice

• Implement real-time failure detection for AI agents.
• Explore policy levers to boost external assurance demand.
• Develop process standards for adaptable AI evaluations.

Topics

• AI Assurance
• AI Standards
• Frontier AI Models
• Agentic AI Systems
• Post-Deployment Monitoring

Best for: CTO, VP of Engineering/Data, Executive, Policy Maker, AI Ethicist, Director of AI/ML

Related on AIssential

• Welcome to the “find out” stage of AI — AI is transitioning from experimental novelty to a phase demanding trust, reliability, and clear business value.
• Building Momentum: From Delhi to Geneva — Global AI governance requires diverse stakeholder engagement, standardized monitoring, and equitable ownership.
• Advance Trustworthy AI and ML, and Identify Best Practices for Scaling AI — US government agencies prioritize trustworthy AI for risk mitigation and best practices for scalable AI implementation.
• Building Trustworthy AI for Enterprise Workflows - with Amar Akshat of PaySafe — Enterprise AI requires embedded guardrails and auditable agentic decisions to bridge the consistency gap between demos and production.
• Prioritizing AI Assurance and Civil Society Engagement Following India’s AI Impact Summit — Effective global AI governance requires civil society participation and a robust, inclusive AI assurance ecosystem.
• 😺 Someone firebombed Sam Altman's house — A widening perception gap between AI experts and the public fuels distrust, alongside emerging AI-driven risks and disruptions.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Partnership on AI.