No one seems interested in learning where AI agents get their data from.

IMO, it's a problem when the data source is unverifiable for agents that move money.
There are already cases where agents acted on false data that brought real damages:
→ Trading bots were wiped mid-execution because their data went stale
→ Autonomous research agents faking their own results
There's barely any infrastructure built for this yet, but some protocols are aiming in the right direction.
For example, @WalrusProtocol lets you prove what data an agent used and that it hadn't been modified at the time of execution. That's the core verification layer that has been missing.
Institutions and AI agents can access decentralized storage at much cheaper costs, which is important because agent-scale verification only works if it's cheap enough to run continuously.
Every failure case so far has been a preview of what happens when the data verification problem stays unsolved at scale.