Responsible AI
How our AI works, and what it will never do
Hiring AI is treated as high-risk by regulators, and rightly so. We designed ours to explain itself, to leave out who people are, and to leave every decision to a person.
What happens when someone applies
- Your team sets the bar. You list the job’s must-haves and nice-to-haves. The AI checks against those, nothing else.
- Personal details come out. We extract the resume text and remove the name, email, phone number and links before the AI sees it. Self-identification answers are never sent.
- Each requirement gets a verdict. Met, partly met or missing, with an exact quote from the resume as evidence.
- We check the evidence. Our code confirms each quote really appears in the resume. If it doesn’t, the quote is dropped and “met” becomes “partly met”, so the AI can’t make things up.
- The score is calculated, not guessed. Must-haves count double. The number comes from the verdicts, so you can always see why.
- A person decides. The score helps you sort. Moving, rejecting and hiring are always done by a person, and recorded with their name.
Our commitments
No automatic rejections
Not with AI scores and not with knockout questions, which only flag an application for a closer look. You can’t set one up, even if you want to.
Candidates are told
The application form explains how AI is used before anyone applies. The version of the notice each candidate saw is recorded.
A way to opt out
Where the law requires it (New York City today), and wherever you choose, candidates can ask for a review without AI. They’re clearly labelled, so they aren’t overlooked.
Fairness you can measure
The adverse impact report compares selection rates across groups, including for high AI scores, and flags anything under the four-fifths rule.
Every score can be explained
Each result is stored with the model and prompt version that produced it, alongside what the person deciding could see.
Your data isn’t used for training
Resumes are sent to our AI providers (Anthropic and OpenAI) only to screen them, under agreements that forbid training on them and with zero data retention, so they aren’t kept after the answer comes back. Before launch, and whenever the prompt or model changes, we test screening on resumes that differ only in names and personal details.
Interviewers stay independent
Interviewers don’t see AI scores, and can’t see anyone else’s feedback until they’ve given their own.
No AI where it doesn’t belong
We don’t analyse video, voices or faces, and we don’t infer personality or emotions.
Questions about our AI
Which model do you use?
Claude, from Anthropic, or OpenAI’s GPT models, each under a no-training, zero-retention agreement. Every score records which model produced it. When neither is configured (for example in a demo), a simple keyword matcher runs instead, clearly labelled “Demo scoring”. Filling in the application form from a resume uses our own reader, not AI.
Does AI see protected characteristics?
No self-identification data is ever sent. Names and contact details are removed before screening, and the AI is told to ignore anything that suggests a protected characteristic, such as age, gender or ethnicity.
Does this make us compliant with NYC Local Law 144 or the EU AI Act?
It gives you the tools: notices, an opt-out, a decision log, and the data for a bias audit. Compliance also depends on how you hire, and some laws require an independent audit. Talk to your employment lawyer.
What about the AI work simulations?
They’re optional and come from our partner hiAIre. Candidates do a realistic task using AI tools, and hiAIre scores how they work. The same rules apply: a person makes every decision.