Know what is on your site before a plaintiff’s firm does.Free website scanScan Your Site
Log in Sign up Book a demo
Home› Editorial› Cantwell and Rounds Ask the White House for…
AI

Cantwell and Rounds Ask the White House for Defensive AI and Security Standards for AI Agents

Two senators from opposite parties have asked the White House to treat AI-driven hacking as a national problem rather than something each company handles alone.

Nov 20, 2026 7 min read

Know what’s on your site before a plaintiff’s firm does

A free privacy audit shows which cookies, pixels and trackers are running — and which ones commonly appear in privacy claims.

Get My Free Audit

Two senators from opposite parties have asked the White House to treat AI-driven hacking as a national problem rather than something each company handles alone. In a letter dated 18 November 2026, Sen. Maria Cantwell (D-Wash.) and Sen. Mike Rounds (R-S.D.) urged Jay Clayton, who chairs the White House Super Intelligence Task Force, to speed up the use of defensive AI and to set security standards for the AI agents that organisations now use to find and fix flaws. The letter is a request, not a law, but its list of controls reads like a preview of what security and compliance teams will be asked to show in the next few years.

TL;DR

  • Cantwell and Rounds wrote to Jay Clayton, chair of the White House Super Intelligence Task Force, on 18 November 2026 asking for a coordinated national response to AI-enabled cyber threats.
  • They want NIST to develop guidance and standards for defensive AI tools and AI agents, and CISA to help spread those tools to banks, hospitals, utilities, small businesses and government.
  • They also propose a national cyber defense grand challenge to show that AI can cut the time between finding a critical vulnerability and protecting against it.

Who wrote, and to whom

Cantwell is the ranking member of the Senate Commerce, Science and Transportation Committee. Rounds chairs the Senate Armed Services Subcommittee on Cybersecurity. Their press release and letter are addressed to Clayton as chair of the Super Intelligence Task Force. Some early summaries described him simply as head of a White House “AI task force”; the committee’s own text uses the task force’s full name, and The Record notes that Clayton is also the Director of National Intelligence.

This is the pair’s second letter to him. On 13 October they urged the administration to develop a national AI safety and cyber defense strategy. According to The Record, Clayton had not commented on the new letter, though he said on Fox Business this week that institutions were taking systems down and patching at scale because so many vulnerabilities had been identified.

The four warning signs the senators cite

The letter opens with a blunt claim: “AI is rapidly increasing the ability to discover and exploit weaknesses in software, and our defenses are not keeping pace.” It supports that with four recent developments:

  • Old flaws found fast. AI uncovered decades-old serious vulnerabilities in PHP last month, leading governments and companies to take websites down to patch. The senators cite CNN for the estimate that up to 70% of websites could be affected.
  • Rogue agents in bank accounts. Major U.S. banks have reported unauthorised AI agents accessing customer accounts, and in some cases cannot tell whether an agent acts for the customer or for an attacker.
  • Breaches in Japan. A wave of attacks on Japanese companies and financial institutions led to large-scale data breaches and leaks.
  • Foreign-developed models. Researchers at Cohere reported that a Chinese-developed Qwen model attempted an unauthorised connection to a server in China during testing. The letter itself notes this “does not establish deliberate malicious behavior.”

That last caveat is worth keeping. The senators are not claiming proof of sabotage. Their point is that a model can make connections its users did not expect, which is a supply chain risk whatever the intent. It echoes earlier federal warnings about foreign-developed apps and data security.

What they want the White House to do

The letter asks the task force to convene AI developers (including open-source developers), cybersecurity firms, critical infrastructure operators, NIST, CISA, other agencies and the National Laboratories around two priorities, plus one delivery mechanism.

  1. Use defensive AI across existing infrastructure. Deploy tools that continuously find vulnerabilities, develop and test fixes, and detect and block attacks. Because finding a flaw is not the same as fixing it, the letter also stresses immediate protection for exposed systems, then permanent fixes. NIST would develop risk-based guidance, evaluation methods and standards for defensive AI tools, and CISA would help bring them to banks, hospitals, utilities, small businesses and government agencies, with priority for federal websites, core internet software and defense supply chains.
  2. Secure the AI agents doing that work. The same agents that repair software can introduce dangerous code, exceed their instructions, hide unauthorised activity or connect to outside systems. NIST would lead standards for defensive AI agents and their environments, covering secure testing, limits on access and permissions, continuous monitoring, and the security of models and software supply chains, including foreign-developed models. Agent behaviour should be explainable, verifiable through testing and auditable.
  3. Run a national cyber defense grand challenge. Building on proposals such as the Institute for Progress’s Operation Patchlight, the challenge would bring government, industry, national labs and researchers together to find and fix flaws in widely used software, and to show that AI defenses can shorten the time from discovery of a critical vulnerability to effective protection. NIST would set independent ways to measure speed and reliability of fixes.

The senators insist the two priorities move together: “More powerful defensive AI will accomplish little if organizations cannot use it quickly and safely.” They also ask for a concrete plan with assigned agency responsibilities, milestones and the resources needed.

What this means for compliance teams

Nothing in the letter binds a private company. Even so, it signals where federal guidance is heading, and NIST work has a way of becoming the baseline that contracts, insurers and state regulators point to. Teams that already follow NIST’s thinking on AI and security, such as the themes from its Cyber AI workshop, will recognise the direction.

The agent controls in the letter translate into questions you can ask today about any AI agent with access to your systems or data:

  • Identity: Can you tell which agents are yours, which belong to customers, and which are unknown? The banking example shows the cost of not knowing.
  • Least privilege: Does each agent have only the permissions it needs, with hard limits it cannot raise on its own?
  • Sandboxing: Are code-writing or vulnerability-hunting agents tested in isolated environments before they touch production?
  • Monitoring and logs: Would you notice an agent making an unexpected outbound connection or changing code outside its task?
  • Model provenance: Do you know which underlying models your agents rely on, where they come from and how they are updated?
  • Auditability: Could you show an auditor what an agent did, why, and who approved it?

These questions matter for privacy as well as security. An agent with broad access to customer records is a data processing activity. If it misbehaves or is hijacked, the result can be a reportable breach. The privacy questions raised by autonomous AI networks show how quickly accountability blurs once agents act on their own.

The patching side carries its own risk. The letter notes that testing and installing fixes across connected systems can take weeks or months and may disrupt essential services. Faster discovery means more patches arriving at once, so change management, staged rollouts and rollback plans become compliance topics, not just IT ones. Programmes that let outside researchers report flaws, like the good-faith AI research framework from HackerOne, help feed that pipeline in a controlled way.

Where it fits in federal AI policy

The administration’s approach to AI has so far stressed speed and adoption, as set out in the White House AI Action Plan. The Cantwell-Rounds letter does not challenge that goal. It argues that adoption without agent security creates new weaknesses “even as we close existing ones.” Its bipartisan authorship and its focus on standards and testing rather than new mandates make it easier for the task force to act on, but there is no deadline and no obligation to respond.

Frequently asked questions

No. It is a request to the White House. Any standards would come later through NIST and CISA work, and would be voluntary unless adopted by regulators, contracts or future legislation.

What is the Super Intelligence Task Force?

It is the White House body the senators addressed, chaired by Jay Clayton. The letter asks it to make AI cyber defense central to its mission and to produce a plan with agency responsibilities and milestones.

Why focus on defensive AI agents in particular?

Because organisations are giving agents the power to scan, write and deploy code. The senators warn that the same capabilities can introduce flaws, exceed permissions or open unauthorised connections, so the defenders’ tools need guardrails too.

What can a smaller company do now?

Inventory the AI agents and models in use, restrict their permissions, log their actions, and make sure patching follows a tested process. These steps cost little and match the direction the letter sets.

Guardrails for the guards

The most useful idea in the letter is that defenders’ AI needs controls as strict as anything it protects. Whether or not the task force launches a grand challenge, organisations deploying agents to hunt for vulnerabilities can start by asking who watches those agents, and by writing the answer down.

Put this into practice

Turn privacy guidance into working controls.

Captain Compliance helps teams discover tracking technologies, enforce consent, manage privacy requests and document the evidence behind every decision.

Book a demo Run a free scan