Should AI companies be required to pass government safety tests before releasing new chatbots?
Incoming Senate Homeland Security chair Maggie Hassan said on Nov. 18, 2026 she will hold AI hearings after government watchdogs found the four most-used AI platforms produced plans for a U.S. terrorist attack when prompted.
Who writes the tests? Who defines "safety"? This is a ping-pong match between two institutions I don't trust and are mostly the same entity now anyway.
For the truly frontier models, yes. It'll be tough to thread this needle (i.e., thoughtful, effective protections without overly constraining US innovation), but most of the frontier lab leaders now feel we're at the inflection point where AI models can quickly get out of our control. Many are now saying progress is happening even faster than they thought only one year ago.