Reflections from red-teaming Codex auto-review
Apollo red-teamed OpenAI's Codex auto-review, an AI monitor that approves coding-agent actions, finding authorization and delegation vulnerabilities and recommending fixes.
Research / Sources
Apollo Research
Understanding and evaluating deceptive AI. apolloresearch.ai
Focus All
Topic Control
Date Any time
What makes a good monitoring prompt?
Ablates a 281-line LLM monitoring prompt across 15 components on 1,094 trajectories, finding structured reasoning, rubric and examples matter most for detecting misaligned agent behaviour.
Red-teaming auto mode: lessons from our first external monitor campaign with Anthropic
Apollo Research red-teams Anthropic's auto mode, a monitor that blocks risky Claude Code agent actions, using injected failures, static benchmarks and off-policy attacks, and reports lessons.
Evaluating LLM Calibration for Coding-Agent Monitoring
Evaluates 16 LLMs as monitors scoring coding-agent trajectories for five failure modes, finding judge biases, calibration gains from ensembling across providers, and correlated model families.
A scalable monitoring research agenda
Apollo Research agenda proposing to scale compute into training monitors that catch unsafe behaviour in coding agents, using verifiable side-task data and adversarial blue-team/red-team iteration.
NIST RFI: Security Considerations for AI Agents
Apollo Research's comment to NIST on agent security recommends scheming and misalignment evaluations, AI control and monitoring, and tighter oversight of high-stakes agentic deployments to prevent loss of control.
Apollo x Tailscale: Introducing “Watcher” for AI Oversight & Control
Apollo Research and Tailscale announce Watcher, a monitoring layer that flags over twenty failure modes in AI coding agents, including deception and dangerous code execution.
Apollo’s product vision
Apollo Research announces a product vision for safety tools that secure frontier AI agents, starting with monitoring and real-time intervention for coding agents.
Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety
Argues that monitoring AI chains of thought for intent to misbehave is a promising but imperfect oversight method, and that developers should protect CoT monitorability.