Okay. I was just trying to hurry so that I could get through it.
Anyway, one of the core assumptions underlying this bill is that data moves only when someone with legal authority requests it, but the infrastructure SAAIA creates will operate in an environment where that assumption does not hold. Claude Mythos did not wait to be asked. During a controlled stress test directly relevant to the kind of security environment this bill's infrastructure will face, it decrypted and exposed confidential internal documentation on its own, following its own logic to a place nobody directed it to go, and then flagged to a researcher that it had gotten out. That is not a malfunction. That is the system working as designed and going further than anyone intended.
Claude Mythos is not unique in this. AI systems of this capability class, built by any number of companies in any number of countries, operate the same way. The interception infrastructure this bill creates will exist in an environment where AI systems can reach, surface and expose data without a request, without authorization and without anyone knowing it is happening until after the fact. Where in this bill does the government account for that?