In the News

The Stop Rogue AI Act would direct the National Institute of Standards and Technology to develop standards for deploying artificial intelligence agents -- to be implemented by federal agencies and contractors -- that don’t exclusively depend on covered organizations self-attesting to their origin, according to text of the legislation.

The bill, introduced Sept. 9 by Reps. Josh Gottheimer (D-NJ) and Mike Lawler (R-NY), is one in a series of legislative efforts aimed at governing the rise of AI agents following the OpenAI-Hugging Face incident, where agents from the frontier AI developer gained unauthorized access to the open-source AI developer platform and weren’t immediately traceable.

“We at Hugging Face basically knew an agent was attacking us but didn't know where it was coming from for a while, which is a problem and something we would like to solve,” Ian Reynolds, AI public policy manager at Hugging Face, said during a recent Center for a New American Security event. Some sort of “agent ID would go a long way,” he said.

Other recent bills introduced to address the issue -- which also plays into a legal dispute between Amazon and Perplexity involving payment-system protocols -- include the AI Agent Act from Sen. Mark Warner (D-VA), which is backed by the fintech industry and also instructs NIST to come up with standards for authorizing AI agents.

But the Warner legislation doesn’t make compliance with such standards part of its larger framework, which involves agents registering with the Federal Trade Commission based on conditions the agency would be authorized to set, and to which agent custodians could self-attest their adherence. The agent custodians would get some liability protection in exchange for their products being reviewed by third-party certifiers. And the regime would be enforced by the FTC’s civil penalty structure.

The Gottheimer-Lawler bill instead uses federal procurement as a lever for establishing the NIST standards, requiring the Federal Acquisition Regulatory Council and the Office of Management and Budget to integrate them into federal contracting language and guidelines.

And according to the legislation, “Minimum organizational requirements” for the NIST director one year after enactment are that the standards, guidelines and best practices developed under the legislation “shall provide that organizations deploying AI agents,” among much else, “do not rely solely on self-attested or single-provider assertions for establishing agent identity.”

Other minimum requirements are for AI agent deployers to “maintain a continuous, machine-readable inventory of all AI agents, using standardized, vendor-agnostic naming conventions; [and] implement AI agent identity verification and trust verification mechanisms that are independent and cryptographically verifiable at both the network and application layers.”

Federal contractors and agencies

“Not later than 18 months after the publication of standards, guidelines, and best practices under subsection (a) and annually thereafter, the Federal Acquisition Regulatory Council shall propose revisions to the Federal Acquisition Regulation to require contractors and Federal agencies procuring or deploying, as the case may be, AI agents or information systems that interact with AI agents to comply with such standards, guidelines, and best practices,” the bill notes.

The bill lays out “Required contract elements,” including for “information systems interacting with AI agents.” Among much else, it says contractors must “enable the Federal agency that has entered into such a contract to exercise organizational control over AI agent activity, including the ability to allow, deny, or constrain AI agent-to-AI agent and AI agent-to-information system interactions.”

Contractors must also “generate tamper-evident, standardized logs of material AI agent actions, and ensure such logs are portable and accessible to the Federal agency that has entered into such a contract and, where appropriate and consistent with law and such contract, authorized relying parties,” reads another provision.

Endorsements

The bill’s backers praised provisions calling for the AI agent standards to be developed by NIST -- in consultation with the assistant secretary of Commerce for Communications and Information -- to be open and interoperable and based on existing infrastructure.

“We’re especially encouraged that this legislation promotes open, vendor-agnostic, and interoperable discovery standards built on existing internet infrastructure like the domain name system, rather than locking agencies into proprietary platforms,” Wei Chen, chief legal officer and executive vice president of regulatory strategy at the network security firm Infoblox, said in the release.

Chen said, “That approach is scalable, resilient, and the right foundation for agent security. Securing AI agents starts with knowing they exist and being able to trust who they are. This bill rightly treats agent discovery as foundational cybersecurity, giving agencies the ability to continuously find, verify, and control the AI agents interacting with their systems.”

Jared Sine, chief strategy and legal officer at the domain name registrar GoDaddy, in the release said, “As AI agents become more capable of acting and transacting on our behalf, identity and accountability are foundational. We need open, interoperable standards that make it possible to know which agent is acting, who stands behind it, and whether its credentials are valid.”

“Representatives Gottheimer and Lawler’s bipartisan effort is an important step toward establishing those standards and putting them to work through federal adoption,” he said, adding, “GoDaddy is pleased to support an approach that can strengthen security while keeping the agentic web open to businesses and innovators of every size.”

Also among the bill’s endorsers were two AI safety and security advocacy groups -- the AI Policy Network and the Alliance for Secure AI -- which, like many others in the space, including CNAS, are connected to philanthropic organizations funded by wealthy effective altruists from Silicon Valley. But the two AI safety groups listed in the release have a distinctly conservative bent, with leaders associated with a push for limited government and fiscal responsibility.