If you are using AI to develop your next big idea, business model or product strategy, there is another question worth asking: what happens to your idea after you type it into the box?
AI has rapidly become the entrepreneur’s brainstorming partner.
We use it to challenge ideas, refine strategies, write code, test propositions and sometimes reveal commercially sensitive thinking we would once have shared only with colleagues, advisers or investors.
But a 2025 research preprint raises an uncomfortable question about that relationship.
Black Box Absorption
In Black Box Absorption: LLMs Undermining Innovative Ideas, researcher Wenjun Cao proposes the concept of “Black Box Absorption”.
The argument is that when creators interact with opaque AI platforms, novel concepts could potentially be internalised, generalised or repurposed through the wider systems surrounding those models.
The paper calls the transferable functional core of an innovation an “idea unit”.
Think about what that could mean in practice.
→ You are developing a breakthrough product strategy.
→ You ask an AI system to stress-test your unique business model.
→ You paste in proprietary code to improve it.
→ You explain the insight that differentiates your company from its competitors.
The concern is not simply what the AI tells you.
It is what happens to what you tell the AI.
The Asymmetry
This is where the paper raises a much bigger issue about the economics of AI.
An individual entrepreneur might contribute one highly valuable idea.
A large AI platform potentially sits across interactions with millions of individuals, companies, researchers, developers and creators.
That creates an important asymmetry.
You know what you typed into the system. You may know far less about what happens to that information afterwards.
Different AI services, subscriptions and enterprise products have different policies governing data retention, model training and privacy. So this is not an argument that every prompt automatically becomes training data.
It is an argument for understanding the terms of the system you are using before putting valuable intellectual assets into it.
In Plain Sight
There is an interesting paradox emerging.
AI can help us innovate faster while simultaneously encouraging us to disclose more of the thinking behind that innovation.
For entrepreneurs, researchers and creators, the question therefore becomes:
Where should the AI assistant stop?
A useful rule might be to separate the work AI can accelerate from the knowledge that gives you your competitive advantage.
Use AI to research.
Use it to structure.
Use it to challenge assumptions.
Use it to accelerate execution.
But think carefully before uploading the secret sauce.
Your unpublished research, proprietary processes, commercially sensitive data, novel algorithms or genuinely differentiating strategic insights may deserve a different level of protection.
Idea Safety Could Become the Next AI Literacy
We already talk about AI safety, cybersecurity, data protection and responsible AI.
Perhaps entrepreneurs also need to start thinking about idea safety.
Before sharing sensitive material with an AI service, understand its privacy controls, retention policies and whether your data may be used to improve models.
For organisations handling particularly valuable intellectual property, consumer AI tools may not always be the appropriate environment at all.
The lesson is not “stop using AI.”
It is “know what you are giving it.”
The Bigger Picture
There was an old warning about the internet economy:
“If you’re not paying for the product, you are the product.”
Generative AI complicates that bargain.
We now pay some AI companies for extraordinary access to intelligence and productivity.
But payment alone does not answer the questions of what happens to our prompts, our data or potentially our ideas.
Perhaps the updated question for the AI age is therefore:
Even when you’re paying for the product, how much of the product are you helping to build?