In theory, a very large staker could run a large share of the network, exit (either normally or get slashed), then rewind the chain prior to the exit, run it on their own independently, and then pop up after a while with a competing fork. This makes syncing from the beacon chain's genesis risky because you could end up following the attacker's fork.

To make up for this risk, Ethereum then relies on distributing recent 'weak subjectivity' snapshots (through other known nodes, block exploers, baked into client releases, etc) to make sure new folks can join the legitimate network and ignore an attacker's. Those snapshots basically rely on social legitimacy to help folks get going with the 'legitimate' chain. It is a trust assumption.

In contrast (at least in theory), in a proof of work network you can 'objectively' determine which of competing forks of a chain is the legitimate one by a simple metric, the 'longest chain with the highest difficulty'. The presumption made here though is that you will have an open internet and honest client software that will not censor the legitimate chain. A trust assumption is still made that those that introduce you to the p2p network aren't hiding a longer chain from you.