Buried in the technical documentation of nearly every major AI system released in the past two years is a phrase most people scroll past: "training data not disclosed." The companies building the most powerful technology in human history have decided you don't need to know what they fed it.
It's the central question of our era dressed up as a footnote.
Researchers have repeatedly found that large language models trained on internet data absorb the biases, assumptions, and blind spots of whatever they consumed.
Study after study shows these systems can replicate racial and gender stereotypes, generate confident falsehoods, and mirror the political leanings of their training sources.
Yet the actual datasets remain trade secrets.
The federal government has noticed — sort of.
The EU passed the AI Act requiring some transparency.
American lawmakers have held hearings, issued frameworks, and produced roughly zero binding rules on data disclosure.
Meanwhile, the same companies lobbying against transparency are the ones selling AI to schools, hospitals, police departments, and the military.
Here's where the dots connect in a way that should unsettle anyone paying attention.
The U.S. government itself is now one of the largest buyers of AI systems.
Palantir, Anduril, and a handful of defense contractors have built businesses on government AI contracts.
These systems make recommendations about surveillance, targeting, and resource allocation.
When the training data is proprietary, the public cannot audit the logic behind decisions that affect real people's lives.
We lived through a version of this before.
Social media algorithms shaped elections, amplified extremism, and rewired public discourse — and we only understood the damage years later, after the platforms had already scaled.
We're now watching the same movie with higher stakes and better marketing.
AI companies have quietly scraped books, art, journalism, and personal data to build products worth billions.
Artists, writers, and ordinary internet users whose posts became training material received nothing — no consent, no credit, no compensation.
Several lawsuits are moving through courts right now, and the outcomes could reshape who owns the building blocks of machine intelligence.
The uncomfortable truth is that we've outsourced our collective memory to systems we can't inspect.
Every ChatGPT response, every recommendation, every automated decision carries the fingerprints of data nobody outside a boardroom has seen.
It means the question of *who controls the inputs* is the real power struggle of this decade — and it's happening while most of us argue about whether the robots will take our jobs. **The takeaway:** Concentration of AI power without transparency is not innovation; it's a quiet transfer of influence from the public to a handful of private entities.
Demand disclosure before deployment, not after the damage.
Final Thoughts
The algorithms will only be as accountable as the people who insist on it.