Embedded assessments provide deeper insight into ai developer practices
Embedded Assessments for Frontier AI
Computers and Society
Summary
It's important to understand how AI companies manage their models internally to ensure safety and reduce risks. The authors discuss a method called embedded assessments, where independent reviewers get employee-like access to internal systems, staff, and documents. This access can help evaluate risks more effectively than just testing AI from the outside. The authors suggest starting these assessments now with continuous monitoring and regular reporting to better manage internal AI use and alignment.
What this means in practice
- •For ai safety teams: Implement continuous embedded assessments to monitor internal AI alignment and security controls for better risk management.
- •For corporate compliance officers: Use embedded assessment frameworks to ensure AI developer practices meet organizational safety and governance standards.
A position paper. It proposes an approach and reports no results.
Authors
Jacob Charnock, Sophie Williams, Zaheed Kara, Markus Anderljung, Alejandro Tlaie Boria, Stephen Casper, Anka Reuel, Jonas Freund
Abstract
Third-party evaluations for frontier AI have mostly tested models through external interfaces before deployment. But the risks from frontier AI models depend on how their developers use and govern them internally. Recently, CEOs of frontier AI companies have committed to hosting embedded assessments. These assessments would give independent evaluators employee-like access to a developer's internal systems, staff, and documentation. First, we argue that this can enable deeper and more flexible assessments of risks that depend on internal systems and practices, while providing access under stronger security controls. Then, we examine seven design questions about scope, information gathering, duration, timing, terms of engagement, disclosure, and escalation. We recommend that frontier AI developers begin hosting embedded assessments now, covering at least three areas central to managing risks from internal AI use: internal agent monitoring, internal agent security controls and permissions, and model alignment. To enable meaningful third-party scrutiny, assessments should be continuous, evaluators should publish detailed reports at least quarterly, and clear escalation mechanisms should be established. These recommendations are intended as a starting point, with further steps needed to realize the full potential of embedded assessments.