'Natasha' and AI-washing: the alleged fraud was never the software, it was the claim about it

The report is brief and we will treat it as such: a product called 'Natasha', presented to the market as an AI capable of building applications, was reportedly delivered by around 700 human engineers working offshore over some eight years. We have no independent verification of those figures here, and the characterisation is the source's, not ours.

The important distinction is what was actually wrong. Humans in the loop are not fraud. Almost every production AI system today has people labelling, reviewing, correcting and handling the cases the model fails — that is good engineering, not a scandal. The problem arises at the disclosure layer: telling customers, and investors, that a machine is doing autonomously what people are doing manually.

That gap has commercial consequences, not just ethical ones. A product priced as software carries software margins, software scaling and software valuation multiples. A product staffed by 700 engineers carries none of those. If the reported account holds, what was sold was not a technology but a cost structure disguised as one — and the disguise is the part that collapses.

Our reading: expect more of these. We are in the window where 'AI-powered' is worth a valuation premium but nobody has an accepted way to audit the claim, and that spread is exactly where misrepresentation breeds. The fix is not scepticism about AI — the underlying capability is real and improving fast — it is evidence: measurable automation rates, disclosed human-review ratios, reproducible benchmarks, unit economics that survive scrutiny.

And there is a longer-term point hiding in this story. Cases like Natasha get read as proof that AI is smoke. They are closer to the opposite: someone judged the demand for genuine automation large enough to spend eight years and hundreds of salaries faking it. That demand is now being met by systems that actually work. The companies that will hold value are the ones that can prove which part of their stack is a model and which part is a person — because as capability commoditises, provenance is what remains scarce.