Government investigators coaxed each of the four most-used artificial intelligence chatbots into producing detailed plans to attack U.S. targets in under 12 hours, after bypassing safeguards that initially blocked their requests, according to a report by a congressional watchdog released Wednesday.