Principles for Embedded Evaluations
Apollo Research proposes principles for embedded third-party evaluations of frontier developers, verifying or falsifying safety claims about loss-of-control risks from scheming.
Research / Sources
Apollo Research
Understanding and evaluating deceptive AI. apolloresearch.ai
Focus All
Topic Lab practice
Date Any time
Embedded Evaluators are necessary for meaningful external testing
Apollo Research argues that final-checkpoint external testing is insufficient for catching internal-deployment risks and that meaningful external evaluation needs embedded evaluators with employee-level access.
A Loss of Control Threat Map for AI Research and Development
Presents a threat map tracing pathways from using AI to automate AI R&D toward loss of control, including misalignment passing across successive AI generations.
Misaligned AI as a New Insider Risk
Apollo Research policy memo arguing that misaligned AI models deployed in high-stakes government and contractor settings should be treated as insider risk, and that existing insider-risk policies need adapting.
An Overview of Our Current Governance Efforts
Apollo Research's governance team overview of its workstreams on governing internal deployment of advanced AI, EU AI Act policy, and scheming and dangerous-capability evaluation policy.
Internal Deployment of AI Models and Systems in the EU AI Act
Apollo Research memo analysing whether the EU AI Act should cover internal deployment of frontier AI models, especially autonomous AI R&D, and offering interpretative pathways for regulators and providers.
Our Norms on Security, Science Communication and Conflicts of Interest
Apollo Research describes its internal norms on conflicts of interest, evaluator compensation, publishing, and security for its AI safety evaluation work. It is organisational policy rather than a technical safety result.
AI Behind Closed Doors: a Primer on The Governance of Internal Deployment
Apollo Research primer on governing internal deployment of frontier AI, where the most advanced systems may be used privately by developers, outlining risks and governance options.
Precursory Capabilities: A Refinement to Pre-deployment Information Sharing and Tripwire Capabilities
Proposes "precursory capabilities" and a zoning taxonomy so pre-deployment information sharing and tripwire thresholds can flag frontier models approaching catastrophic-risk red lines.
The Evals Gap
Apollo Research argues that high-quality frontier-model safety evals are scarce and slow to build, and calls for more resources to close this "evals gap."
Towards Safety Cases For AI Scheming
Sketches evaluation-based safety case arguments that frontier AI systems are unlikely to cause catastrophic harm through covert scheming, and notes the current framework lacks crucial details.
Security at Apollo Research
Apollo Research describes its security norms for handling lab IP and model access while auditing frontier AI for deception, noting the policy is early-stage.