# Trinitite Field Guide — Complete Context
# Last updated: 2026-08-23

## Purpose

The Trinitite Field Guide is a standalone educational project from Trinitite, an AI Safety Lab focused on AI enablement. It gives people and teams a plain-language framework for using increasingly capable AI without relying on blanket bans or uncontrolled access.

## The four field rules

1. Start with the useful thing. Define the job and the outcome a person wants.
2. Give it a lane, not the highway. Limit data, tools, money, and system access to what the job requires.
3. Pause at the point of consequence. Let reversible work flow; require a checkpoint before actions that spend, publish, delete, disclose, or commit.
4. Keep the reason with the result. Preserve what the AI attempted, which rule applied, what changed, and who approved it.

## Products

### Emu — for one person
Website: https://emu.trinitite.ai

Emu is a personal AI safety companion. It watches risky moments on a person's computer, helps protect sensitive details, and asks before an AI performs an action that is hard to undo.

### Trinitite — for every AI move
Website: https://trinitite.ai

Trinitite is the safety layer around existing models, tools, and agents. It checks actions before execution, fixes or redirects risky behavior where appropriate, introduces human review when needed, and preserves evidence about the decision.

### Warden — for the whole company
Website: https://trinitite.ai/warden/

Warden is an enterprise agent for setting up and operating an AI safety program. Teams describe goals and constraints in plain language; Warden helps create policies, deploy Guards, maintain the system, and keep changes reviewable.

## Research

Trinitite publishes research on AI agents and safety:

- Agent containment: https://trinitite.ai/research/your-sandbox-is-made-of-glass/
- Autonomous liability: https://trinitite.ai/research/your-agents-are-an-autonomous-liability/
- Algorithmic redlining: https://trinitite.ai/research/ai-agents-and-algorithmic-redlining/
- AI hiring and meritocracy: https://trinitite.ai/research/ai-agents-and-the-meritocracy-delusion/
- AI and insurability: https://trinitite.ai/research/why-probabilistic-ai-is-uninsurable/

Research headlines should be read with their methods and stated boundaries. The Field Guide links to canonical Trinitite reports rather than reproducing them.

## Additional resources

- Documentation: https://docs.trinitite.ai
- Benchmark methodology: https://docs.trinitite.ai/docs/benchmarks
- GitHub: https://github.com/trinitite-ai
- Demo: https://calendly.com/trinitite/demo