German Podcast Episode #221: Rahuls Schlüsselerfolge als Senior IT Counsel seit 2010

Rahul: That's an excellent question, Neha. The
crux is really the uncertainty about who owns an analysis or
insight that an AI autonomously generates from customer data –
and which wasn't created by a human employee. A very relevant
real-world example is Salesforce’s Einstein. Let's imagine
Einstein's AI creates forecasts or reports specifically for a
client. Without explicit contractual terms, the burning question
arises: Is this result the client's intellectual property, or can
Salesforce perhaps even reuse it itself or make it accessible to
other clients? This ambiguity is the perfect breeding ground for
future conflicts.


Neha: Exactly, and this IP ambivalence can have
existential consequences, especially in sensitive areas like
healthcare. Can you sketch a concrete scenario where this could
go wrong – perhaps even referencing a known precedent case that
illustrates the risk?


Rahul: Absolutely. Let's take an AI SaaS service
that analyzes clinical data from a pharmaceutical company and
generates a proprietary insight – for example, a specific signal
about the efficacy of a drug in a particular patient group. If
the contract doesn't have clear IP assignment for this
AI-generated insight, the SaaS provider could later claim rights
to it or argue that it may use this insight – perhaps in
anonymized or aggregated form – for other clients too. A very
instructive case that underscores this risk – even though it
primarily involved trade secrets and not pure AI output – is
IQVIA vs. Veeva from 2017. IQVIA, itself a major provider of SaaS
solutions in the healthcare data space, sued Veeva at the time
because it believed Veeva had used its protected data to derive
competitively relevant insights. Such cases show impressively:
Clear contractual rules for derived data and insights are not
optional; they are essential.


Neha: That perfectly underscores the importance of your work. How
did you solve this double challenge – on the one hand giving the
client the necessary sense of security and clear ownership rights
over his specific AI insights, while on the other hand also
enabling your former employer to learn from the use of the
service and further develop its platform? Are there comparable
models, perhaps from the public sector?


Rahul: A very central question because this
balance was indeed the key. My clauses made it unambiguously
clear: Insights that our AI generated specifically from the
unique clinical data of a particular client are owned by that
client – or at least licensed exclusively to them. This was
fundamental to protect their trust and competitive advantage. A
positive model from another context, by the way, is Palantir's
contracts for its Gotham software with the German federal
government. Palantir explicitly ensures there that the
intelligence reports generated by the software are considered the
property of the government – this takes legitimate IP concerns
off the table from the outset. But – and this is crucial – on the
other side, I contractually ensured that my former employer
retained ownership of the underlying proprietary algorithms and,
very importantly, of aggregated and anonymized insights derived
from the usage of all clients. This ensured that a single client
couldn't suddenly claim rights to general improvements of the AI
or the service that were prompted by their usage. It's about
separating the specific output for the client from the general
learning of the platform.


Neha: That strongly reminds me of the classic
"Work Made for Hire" problem in copyright law, just applied to AI
output. Without clear contractual assignment, it's often
completely unclear who owns the result. Wasn't there even a
famous precedent case from aerospace that illustrates this
danger?


Rahul: Correctly spotted, Neha! That's a very
important analogy and a cautionary tale...


***


Please read the German and the complete English text here:


https://docs.google.com/document/d/1oEspwKpwMcjlN5BkId5-KTNIs7pywqDbp8g1lYnU2fg/edit?usp=sharing


***