Security
We document Aleida with one house rule: never claim a control without being able to show where it lives. This page follows the same rule.
The assessment
Every quote an assessment relies on is checked word for word against what the candidate wrote, and removed if it isn't there. If a run fails its own checks, it is redone.
exchanged the item and offered the customer a coffee while we waitedFound word for word in the answer
I have managed a team of twenty people.Not in the answer, removed
Logging
Every model call is logged with time, model, size and status, not the content.
The conversation
No video, no voice, no images. No tone of voice, no personality scoring and no sentiment analysis.
Human oversight
The write-up is an order in which to talk to candidates, not a decision. You see the whole conversation and can always overrule it.
Access
Candidate data sits behind row-level security with per-company rules. Every request is checked against the company the candidate belongs to.
Models
Aleida trains and fine-tunes no AI model, and none of your data is used for that. The models are called over an API.
Who does what
The employer is the data controller. Aleida is the data processor and handles the data only under our data processing agreement and your instructions.
Candidates can contact the employer to find out what has been stored, or to have it corrected or deleted. The conversation says at the start that it is held by an AI.
The EU AI Act
Aleida is used in hiring, so we assess it as a high-risk system under Annex III 4(a), with ourselves as the provider. The requirements apply from 2 December 2027.
We already document against Articles 9 to 14: risk management, data, technical documentation, logging, transparency and human oversight.
Documentation on request
Want to know more?
Leave your work email and we'll get in touch, usually the same day.