Some warnings about AI enabling bioweapon design sound alarming; others dismiss them as hype that ignores the tacit knowledge needed for real wetlab work. The truth matters enormously for policy. For those who've looked at the evidence seriously, how much does AI actually lower the barrier?