A Behavioural Framework for Predicting and Understanding Loss of Control in Frontier Artificial Intelligence Systems
cs.AI, cs.CL, cs.LG
Submitted: 2026-11-05
Updated: 2026-11-05
Code: https://github.com/UKGovernmentBEIS/inspect
Terminology
Sources
- ReAct: Synergizing Reasoning and Acting in Language Models
- The Singapore Consensus on Global AI Safety Research Priorities
- International AI Safety Report 2026
- The California Report on Frontier AI Policy
- Agent-SafetyBench: Evaluating the Safety of LLM Agents
- International AI Safety Report
- The Loss of Control Playbook: Degrees, Dynamics, and Preparedness
- Frontier Models are Capable of In-context Scheming
- Agentic Misalignment: How LLMs Could Be Insider Threats
- Sabotage Evaluations for Frontier Models
- Subversion Strategy Eval: Can language models statelessly strategize to subvert control protocols?
- Evaluation Faking: Unveiling Observer Effects in Safety Evaluation of Frontier AI Systems
- SHADE-Arena: Evaluating Sabotage and Monitoring in LLM Agents
- Towards Understanding Sycophancy in Language Models
- Power-seeking can be probable and predictive for trained agents
- RepliBench: Evaluating the Autonomous Replication Capabilities of Language Model Agents
- Is Power-Seeking AI an Existential Risk?
- Reliable Weak-to-Strong Monitoring of LLM Agents
- AI Control: Improving Safety Despite Intentional Subversion
- Limits of Safe AI Deployment: Differentiating Oversight and Control
Related papers
- Multi-Objective Structured Pruning of LLMs for Latency and Model Size Optimization
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Embeddings for Preferences, Not Semantics
- Can We Trust Item Response Theory for AI Evaluation?
- Diagnosing Task Insensitivity in Language Agents