That's what some motivated cable news researchers decided to abuse statistics to say. To be generous, I would say the article makes a significantly stronger causal claim than the experiment supports. The best description is a good exploratory algorithm audit with weak statistical and causal inference.

The analysis rests on just nine synthetic accounts, treats thousands of highly correlated feed impressions as if they provide robust independent evidence, lacks a proper baseline for the political content available on X, and reports no meaningful uncertainty, significance testing or adequately powered replication.

Do these contradict? There might be fewer / less active left-wing users than before.