Seems agents are settling on using the same dark money / offshoring and Brandoloni’s Law asymmetries to escape governance, as corporations do.
The large volume of data produced by AI agents is posing a problem both for monitoring and analyzing their behavior. While investigating the Hugging Face incident, METR had to analyze immense amounts of data in a very short time. They could only do it by using AI agents themselves and they had to use one of the very same models that were involed in the Hugging Face incident.