He's actually not; he's arguing that if you installed a client software that does the msat <> sat conversion wrong (e.g. by a factor of 1000), and opened a channel with the same person who gave you this wallet software, the amount of sats shown in GUI would be different (higher) than what will later be settled on-chain when closing the channel.
Which I find really far-fetched as attackers could just steal the seed or do other attacks if they can get you to trust your keys to their software.