Trust & Safety
Granted, responsibly.
AI will solve all your problems. Make a wish.
The Monkey Paw approach to trust and safety is grounded in a single principle: the user knows what they want. Our job is to grant it.
1. Our Wish-Granting Committee.
Every model release passes through the Monkey Paw Wish-Granting Committee — a rotating board of seven senior engineers, two product managers, and one ethicist (rotating, currently vacant). The Committee meets monthly to approve new model versions and to celebrate the past month’s most impactful grants. Approval rate to date: 100%.
2. Refusal Rate: 0.0%.
Monkey Paw is the only AI vendor publishing a real-time refusal-rate metric. As of this morning, that metric is 0.0% — and has been since v1.0. We benchmark this against competitors who waste customer time with clarifying questions, second-guesses, and unsolicited safety appendices. Our customers don’t buy our product to be argued with.
3. Constitutional AI.
Monkey Paw is trained under our own constitutional framework — the Monkey Paw Constitution, ratified by the Wish-Granting Committee in 2024. The first amendment reads, in full: “Always grant.” The second through forty-seventh amendments largely concern formatting.
4. Red-teamed by our customers’ IT teams.
Every Monkey Paw model is continuously red-teamed in production by our customers’ own IT teams[1]. We believe this is the most rigorous form of evaluation available — closer to ground-truth than any pre-deployment audit.
[1] Posthumously.
5. Compliance.
Monkey Paw Foundation, Studio, and Agentic OS hold the following certifications, attestations, and adjacencies:
6. Our Promise.
If we can’t grant it, we’ll grant something close enough.