AI in OSINT, without exposing your own investigations
Cross-check public sources in a few queries, yes, but never at the price of the target’s name or the case’s very existence sitting in a public AI.
Private investigators, journalists, lawyers, HR, compliance and security officers: your research is about people, and every name typed into a consumer AI without a data processing agreement reveals who you are looking into, and why. The search then leaves more traces than its target.
The point is not to give up AI, it sorts and cross-references what hours of browsing cannot cover, but to frame it: lawfulness and proportionality settled before the search opens, documented sources, nothing identifying beyond your walls.
- Hours lost manually cross-referencing scattered, volatile sources
- Target’s name and the case’s existence handed to an AI without a data processing agreement
- Searches opened without a framing note · lawfulness and proportionality judged after the fact
- Screenshots unusable the day the trace has to serve as evidence
See for yourself
A prompt structured in 5 blocks: Instruction · Role and audience · Data · Constraints · Format. An unframed AI invents; the frame stops it.
In a workshop, we build these prompts on your real cases: each sheet sets its legal scope, its chain of evidence and its safeguards, and nothing identifying leaves for a public AI.
Note de cadrage · la légalité et la proportionnalité d’une recherche, avant de l’ouvrir
The lawfulness question asked and settled before the first query, not after the fact.
See it in Chaptal →Vérification d’un candidat avant embauche · le périmètre autorisé
What Swiss law lets you verify about a candidate, and what must stay out of the search.
See it in Chaptal →Chaîne de preuve · capturer une trace en ligne pour qu’elle serve encore dans six mois
The trace captured, timestamped and documented so it remains usable in a file.
See it in Chaptal →30 use cases cover your line of work in the Chaptal catalog, out of 750 in the corpus · each one states what it produces, what it expects from you, and what it refuses to do.
« Le Collaborateur, augmenté »
Your teams learn to frame AI on their real research, safeguards included.
See the training →AI usage audit
A survey of the research already done with AI in your teams, and the framework to hold it.
See audit & advisory →Foire aux questions
Is OSINT research on a person legal in Switzerland?
Consulting public sources is allowed, but gathering information about a person is data processing under the nLPD. Purpose, proportionality and transparency apply · a framing note before the search opens protects the investigator as much as the person searched.
Can we type a person’s name into a public AI to research them?
You would reveal to a tool without a data processing agreement who interests you, and why. Framing means anonymized queries and tools that train no public model on your content.
What can we verify about a candidate before hiring?
Swiss law limits the check to information relevant to the position. The framed prompt writes that scope in black and white · what enters the search, and what must stay out of it.
Is a screenshot enough as evidence?
A screenshot on its own is easily contested. Timestamp, source, context and a register of searches document the trace so it remains usable months later · that is the purpose of the chain of evidence.
What if our own company is the one exposed?
OSINT also works in defense · mapping your public digital footprint, reducing it, and monitoring data leaks that concern you. The sheet covers these defensive protocols alongside the research itself.