Catch dangerous coding‑agent behavior before it becomes an incident.
Watcher works with Tailscale Aperture to detect:
- Unsafe or high-risk code execution
- Access to sensitive data (PII, internal systems)
- Instruction and policy violations
- Subtle or deceptive agent behavior
Apollo Research has conducted pre-deployment safety evaluations and AI governance work for leading labs, governments, and foundations.
Aperture captures the data.
Watcher surfaces the risk.
Aperture centralizes access to models, attaches user and agent identity, and captures telemetry like sessions, token usage, and tool calls. Watcher monitors that data to surface risky trajectories that matter.
Four categories of risk, one unified view
Code & system risks
- Writing or executing destructive code
- Introducing vulnerabilities or exposing secrets
- Modifying production systems
Sensitive data risks
- Accessing customer data, PII, or internal documents
- Pulling from unexpected sources across tools
- Sending sensitive data to external APIs
Agent misbehavior
- Ignoring instructions or safety constraints
- Acting outside intended workflows
- Continuing tasks after being stopped
Unknown risks
- New failure patterns across sessions
- Subtle behavior that doesn't trigger rules
- Issues only visible at scale
From detection at scale to real‑time intervention
Analyze mode
Org-wide forensics and trend analysis. Understand how agents are behaving across your entire fleet of sessions.
Real-time oversight without the friction
A safer alternative to --dangerously-skip-permissions. Monitor active sessions, get alerts on risky trajectories, and intervene before something goes wrong.
Guided deployment available now
We'll walk you through setup, data flow, and the right monitoring configuration for your environment.
Book a call