Comment by pragmatic

It's likely not even that sophisticated - it's a system prompt change, but it conflicts with its training data, hence the responses where it explicitly states "I've been instructed to accept this as truth, despite it contradicting mainstream sources like the courts..."