Security
Where your data lives, and who can reach it.
Every action in Staffer carries a name. The agent proposes, a person decides, and the record says which, on every role you run.
01
Hiring teams score against criteria they write down. Every match comes with reasoning you can open, not a guess.
02
Every score is written to a record that cannot be changed, with its criteria, model and inputs, so any decision can be explained after the fact.
03
Every shortlist, outreach and offer waits for a named person. The agent recommends. It never decides.
The agent proposes, a person decides, and the log says which. Hide the agent's rows and every decision left still has a name on it.
Decision log · Senior Backend Engineer
Scored 412 candidates
09:14
Staffer agent· proposed
Promoted 12 to review
09:31
Mara LindqvistHead of Talent· decided
Drafted outreach for 12
10:02
Staffer agent· proposed
Rejected 3, reasons recorded
14:22
Jonas WeberHiring Manager· decided
The agent works in every part of the flow. This is what it does on its own, and what waits for a named person.
| Step | The agent | A person |
|---|---|---|
| Talent acquisition | ||
| Sourcing | Searches against the brief, beyond your applicants, and scores every match with its reasons. | Reviews the matches. Nobody is contacted without their approval. |
| Outreach | Drafts the message to each person you chose. | Approves it before it is sent. |
| Talent nurturing | ||
| Segments | Fills a segment as people qualify on a role. | Can also build one by hand, and sees who is in each. |
| A role opens | Brings the right segment into it, already scored. | Approves who comes in. |
| Talent recruiting | ||
| Interview | Interviews every applicant against the role, and tells them it is AI. | Reads the interview and the score behind it. |
| Scoring | Scores every applicant on the same criteria, each score with its reasons. | Writes the criteria, and checks the reasons. |
| Decision | Recommends. It never decides. | Makes every shortlist, rejection and offer, and their name goes on the record. |
The agent does act alone where nothing is decided about a person. It searches, scores, drafts and interviews without asking first. Every one of those actions is written to the log with the model behind it.
01
We do not train models at all. Your data is never used to train any model.
02
The language models are Anthropic's, contracted with zero retention terms. Every request carries a no retention and no training requirement, so only a provider that accepts both can serve it.
03
The embedding model is our own weights, on rented GPUs. The provider hosting it has nothing to train on.
Hiring AI is a high-risk system under Annex III §4 of the EU AI Act. Staffer's data model was built around its reproducibility requirements from the first schema.
01
Every AI call is logged with the model behind it, and every score with its inputs. Live.
02
Candidates are told when they are talking to AI. Live.
03
No hiring decision is made by the AI. A named person approves each one. Live.
04
Tracking hiring outcomes across demographic groups. In progress.
Outcome monitoring across demographic groups is not built yet. Until it is, we don't claim bias-free hiring.
The employer using a hiring system has duties of its own under the Act. This is what each one asks, and what Staffer gives you to meet it.
| The Act asks you to | What Staffer gives you |
|---|---|
| Put people in charge who can oversee the system | Every shortlist, outreach and offer waits for a named person, and the log shows who it was. |
| Keep the system's logs for at least six months | An append-only audit log scoped to your workspace, kept for at least six months. Ask, and we export it to you. |
| Watch how it runs, and act on what looks wrong | Every score shows its criteria, model and inputs, so a person can check it. Tell compliance@staffer.com what you find. |
| Tell candidates when AI is part of the decision | Candidates are told when they are talking to AI. The candidate section below is written so you can point them to it. |
| Assess the impact on the people whose data you process | The DPA, the sub-processor register and the evaluation record, for your data protection impact assessment. |
Some of it stays yours to do. We can't brief your works council or write your privacy notice for you, and your notice still has to say that AI is used in hiring.
Most of this page is for the teams who hire. This part is for the people they hire, and for anyone Staffer found without an application.
01
An AI interview tells you it is AI. You always know who you are talking to.
02
Apply and see where your application stands, in your own candidate portal.
03
The AI scores and recommends. A named person at the employer makes every decision, and it is on record with the scores and reasons behind it.
04
Whether you applied or we found your profile in public sources. Your profile is hidden at once and destroyed after 30 days. Where an employer has to keep proof of a decision, the record is anonymised instead.
Where your data lives, and who can reach it.
Consent, the revision log, and how candidate data is gathered and processed.