Whenever someone puts "AI" and "blockchain" in the same sentence, I want to hide. The two technologies have become a kind of buzzword salad — sprinkled over pitch decks to suggest depth and innovation, while quietly meaning nothing.
But lately, I've been watching a few places where the combination genuinely works. Not in the whitepaper way. In the "we shipped this and people use it" way.
The honest problem each one solves
Let me strip this back to first principles.
AI is very good at finding patterns in data and making decisions. It is notoriously bad at explaining why it made a decision, or proving the decision hasn't been tampered with.
Blockchain is very good at creating tamper-evident records that many parties can agree on without trusting one central party. It is notoriously bad at, well, doing anything interesting with the data it records.
Put them next to each other, and you get something one can't do alone: an AI system whose decisions, training data, and audit trail can be verified, not just believed.
Where this actually helps
Supply-chain integrity. A European food producer uses AI to spot inconsistencies in a shipment's provenance data, but records each reading on a shared blockchain that the farmer, processor, shipper, and retailer all see. When something goes wrong, there's no "he-said, she-said." There's a record.
Medical research consortia. Federated AI lets multiple hospitals train shared models without moving patient data off-site. A blockchain keeps the audit trail of who contributed what and when — useful for regulatory compliance and research credit.
Model provenance. As generative AI floods the world with synthetic content, a few teams are using blockchain to sign the outputs of AI systems. Not to stop misuse, but to give downstream systems a way to ask: "was this produced by a model, and which one?"
What doesn't need a blockchain
If you're building an AI product and someone suggests adding a blockchain, ask these three questions first:
- Do multiple untrusting parties need to share this data?
- Does a tamper-evident history genuinely change outcomes?
- Would a well-maintained audit log in a traditional database be enough?
If you answer "no" to the first two and "yes" to the third — which is most of the time — you don't need a blockchain. You need a database and a decent access policy.
The mistake people make isn't that blockchain is bad. It's that they reach for it before they've articulated the trust problem it's supposed to solve.
A small note on energy
The older proof-of-work blockchains are ruinous for the planet. Any serious conversation about AI + blockchain today should involve only proof-of-stake systems or permissioned chains — both of which are orders of magnitude more efficient. If someone's still selling you on Bitcoin-class infrastructure for a supply-chain application, that's a signal.
Where I'd start
If you're genuinely curious about this space, start with a concrete trust problem. Something like: "Our auditors need to verify our model hasn't been tampered with between training and deployment." That's a sentence with a real answer, and the answer might involve blockchain.
The sentence "We're exploring how AI and blockchain could come together" is not that sentence. It's a pitch deck. Don't build on it.
We've helped a few clients figure out whether a blockchain actually belongs in their AI stack (usually: no, occasionally: yes). If you're wrestling with a similar question, come and talk to us. We'll be honest.
