The simplest design, and its single point of failure
The most straightforward way to get external data on-chain is to have one party, an operator, a company, a single server, fetch the data and submit it in a transaction. This design is simple to build, cheap to run, and fast, there's no coordination overhead because there's only one voice. But it inherits an obvious weakness: the entire system's correctness now rests on that single source being honest, competent, and continuously available. If that source is compromised, goes offline, makes an error, or is simply incentivized to lie, every contract depending on it inherits the failure with no built-in check.
This isn't a hypothetical concern, it's the direct consequence of collapsing the oracle's trust model down to one actor. A single centralized feed can be attractive for low-stakes or internal use cases where the cost of being wrong is small, but for anything securing significant value, a single point that can be wrong, hacked, or malicious is generally considered an unacceptable concentration of risk, precisely because it reintroduces the kind of unilateral trust that decentralized systems are built to avoid.