The permissions problem is one of the gaslighting issues I can’t get over. The security bros on here that talk about AI security and governance are liars. Agents by definition can’t be read only programs.