Part two: the private-sector framework

I propose an Agentic Infrastructure Framework. Developers, deployers, security researchers and consumer representatives would write and maintain it. Government recognizes it and sets the floor; it does not draft it. Its six parts consolidate existing industry and standards work, including NIST's agent identity and risk management efforts, CISA's secure-by-default guidance, the open agent protocols and ISO/IEC 42001, into a single form the statute can recognize.

The statute sets conditions for recognition: open membership, published documents, seats for consumer and security representatives, regular review, and compliance with antitrust law. The framework may exceed statutory protections but cannot weaken or waive them. That is what keeps this capitalist and not captured. Companies compete on how well they meet the standard. They do not get to write their own exemption.

Warner's introduced AI AGENT Act likewise pairs recognized certification bodies with a rebuttable compliance presumption and security-related access procedures (S. 5051, section 3(d)(4)) ^[8]; its platform coverage differs from the access right proposed in Part one. The recognized framework would translate existing guidance into auditable requirements; management-system certification alone would not establish that a product is safe or legally compliant ^[5][6].