Safety

Power without
accountability
is a liability.

We build AI that can reason about the world's hardest problems. That kind of capability demands a matching level of rigor about what can go wrong — and how we prevent it.

How we think about it

Last reviewed — June 2026

v2.4.1

Threat model

Four layers. All active.

What the model does

We define explicit behavioral constraints during training — not just as filters on output, but as internalized tendencies. The model should refuse harmful requests because it understands why, not because a regex caught the word.

99.7%

refusal accuracy on red-team evals

Our principles

What we commit to.

Hover each principle to read our reasoning.

I

Understand before deploying

II

Failures are features

III

No single point of trust

IV

Transparency over optics

V

Users are people

System checks

Current status.

All systems nominal
Behavioral eval suite
passing
Red team assessment
passing
Preference drift monitor
passing
Jailbreak detection
passing
Policy compliance layer
passing
External audit
pending

Our commitment

Safetyisnotafeature.
It'sthefoundation.

We publish our safety methodology, our eval results, and our failures. If you find a vulnerability in Caether's systems or models, we want to hear from you — before anyone else does.