Chainlink’s nodes are somewhat decentralized (depending on the distribution of nodes). However, the relatively low number of nodes means that in practice, an attack on several nodes would not be significantly more difficult than an attack on a single central party.
I’d also like to point out that chainlink’s long-term plan is to transition towards using software enclaves (Intel SGX), which is by any measure a really bad idea if you’re planning on running arbitrary user code, which they are. There have been a number of vulnerabilities found in SGX, and the consequences of a successful exploit would be very serious.
It’s concerning to see well-known projects in crypto ignore well-known vulnerabilities when developing. Section B in chainlink’s whitepaper discusses “SGX Trust Assumptions” - that is, that chainlink requires that Intel hardware is not compromised. Even if this were at all a reasonable assumption, the security properties of SGX itself are not discussed - for example, Intel does not consider side-channel attacks as part of SGX’s threat model and as such does not protect against it. These vectors have been successfully exploited.