Over the past couple of years in the crypto space, I’ve become increasingly cautious of projects that claim to be "completely trustless." Honestly, the most practical question when something goes wrong is: who will compensate?

The essence of oracles is to provide a channel for external data to connect to the blockchain. But this process is full of uncertainties—price data can be incorrect, multiple data sources might conflict, and some might even intentionally feed false information. Most projects either pretend these issues don’t exist or try to eliminate uncertainty with complex mechanisms. But in reality, it’s a mess that can’t be fully cleaned up.

I think a more pragmatic approach is: instead of claiming the system is flawless, openly acknowledge the chaos and design mechanisms to manage it. For example, split data processing into two stages—quick judgment and source verification are handled off-chain near the data source, while final decisions involving fund flows are kept on-chain. This way, even if one part fails, the losses are contained.

In DeFi, maintaining trust comes at a cost. Over-collateralization, conservative parameters, centralized backdoors—these are all trust taxes. How to reduce them? Rotate multiple data sources for price feeds, implement multi-layer verification mechanisms, and combine economic incentives and penalties. The core idea is to make malicious gains less profitable than the costs, leaving no room for bad data to hide. That’s practical operation.

Regarding AI applications, I am most cautious. AI as an auxiliary tool for anomaly detection is useful—it can quickly spot suspicious signals. But if you treat it as the final arbiter, you’re introducing a black-box trust layer—no one truly understands how it makes decisions. This goes against the original intention of decentralization. AI can be a helper, but never a decision-maker. Too many projects have been ruined by this mistake.