AI on the job: what information should you share with an assistant?
In Basel, pharmaceuticals, life sciences, Rhine logistics and international R&D rely on information whose circulation needs to be controlled. The management question is specific: how can teams be helped to decide what they may share with an AI assistant before drafting their request?
What the literature establishes
Alessandro Acquisti, Laura Brandimarte and George Loewenstein (2015, Science) review research showing that disclosure decisions depend, among other things, on uncertainty, context and how choices are presented. Privacy preferences therefore do not translate automatically into consistent behaviour. For managers, this is a reason not to equate knowing a rule with applying it at the moment information is shared.
Context changes what people are willing to reveal
Leslie K. John, Alessandro Acquisti and George Loewenstein (2011, Journal of Consumer Research) demonstrate experimentally that the context in which information is collected can change willingness to disclose sensitive information. In their experiments, a less professional-looking setting can elicit more disclosure than a more formal one. These findings suggest a hypothesis to test for AI: a familiar conversational interface could make the boundary between ordinary conversation and data transmission less salient.
The premature conclusion
One might conclude that all internal information should be prohibited in every interaction with AI. Yet these studies neither assess the security of a particular tool nor justify treating all internal information as sensitive personal data. A useful rule must distinguish the information category, the authorised environment and the conditions of use, rather than rely on a general impression of trust or distrust. (our executive and employee training programmes)
What the evidence cannot establish
This research does not examine the current professional use of generative assistants in Basel-based companies. Extending its findings to manufacturing secrets, research results or shipment data remains a working hypothesis, not a demonstrated effect. Likewise, completing an awareness session or signing a policy does not directly measure the ability to recognise an unauthorised disclosure in a concrete situation.
A practical check in Basel
As part of SHR — Swiss Human Resources’ “IA on the job” programme, one possible exercise uses fictional requests drawn from pharmaceuticals, life sciences, Rhine logistics and international R&D, without any real confidential data. Before and after training, each participant specifies what they would share, remove or submit for approval, identifying the intended tool and environment. The main measure is the proportion of cases in which the proposed disclosure breaches a rubric approved in advance by security, data protection and business representatives; unnecessary refusals are recorded separately. Using two comparable sets of cases makes it possible to check progress on the exercise without presenting it as proof of compliance in real situations. To go further: explore the AI on the Job training in Basel, or browse our executive and employee training programmes in Switzerland.
In pictures: AI on the Job in Basel
- ai on the job
- Basel
- research
- training basel