September 21
Everyone Has an AI Policy. Almost Nobody Has Evidence.
AI Assurance
Anyone can ship an AI feature.
Proving it was safe to ship is the hard part.
Keido Labs is an AI research and development lab — mechanistic interpretability, AI psychology, AI safety — with an assurance practice built on that work. We audit LLM features already in production. Independent by design.
Companies shipping LLM features are roughly where SaaS was in 2010. Real exposure, real consequences, and no accepted way to show anyone you took it seriously.
Your customers' security questionnaires have grown an AI section. Your answers haven't.
A prompt injection arriving inside a document nobody inspected. An agent holding permissions someone granted in a hurry. An eval judge that has never once been checked against a human. A chat surface that meets a person in crisis and improvises.
None of it shows up in your analytics. All of it shows up in diligence.
The teams building these products usually care a great deal. What they don't have is an independent read on what they actually built.
What we do
Fixed scope, fixed price, findings you can reproduce. Scoping calls are free and the quote lands within 24 hours.
AI Assurance Audit
£4,000–8,000
Five layers on one live LLM feature: adversarial security testing, the state of your evals, what the conversation does to your users, your governance gaps, and whether you would notice it breaking. Severity-rated findings, reproducible evidence, a remediation plan — and a re-test once you have fixed them.
Psychological Safety Deep Dive
£6,000–12,000
For anything with a chat surface the public can reach. A risk taxonomy built for your product, a judge validated against human raters, a gate you can run in CI, and documentation written in the language the statutes that apply to you actually use.
Assurance Retainer
£1,000–1,500 / month
Because the model changes underneath you. Drift review, a regression run each time a provider updates, incident review when something goes wrong, and a one-page note your board can read.
Full scope, every tier and what it costs•
Not sure yet? The AI Risk Snapshot is a single day, from £1,200.
Someone is going to ask what you did about your AI feature.
Better it is us, on a Tuesday, than a customer's security team mid-deal.
We teach AI to care
July 6
The bench and the judge. Local stack to keep AI psychologically safe.
AI Psychology
June 9
AIPsyBench - the way to keep AI conversations safe
AI Psychology
May 9
What happens when you strip a language model of its feelings
AI Psychology
March 25
Whether, Not Which. What happens inside LLM when it process emotions.
AI Psychology