Two senators want the White House to treat AI agent security as an urgent problem, not a future one. Sen. Maria Cantwell, the top Democrat on the Senate Commerce Committee, and Sen. Mike Rounds, who chairs an Armed Services cybersecurity subcommittee, sent a joint letter urging the administration to act immediately to protect critical infrastructure, financial institutions and widely used software, according to the Senate Commerce Committee.
The letter was sent to Jay Clayton, a senior White House AI official. Coverage disagrees on his exact title. One report calls him chair of the White House Super Intelligence Task Force, while NBC News calls him the White House AI czar. The exact date of the letter was not confirmed in the sources reviewed, but reports from the past few days describe it as recent.
Why the senators say defenses are falling behind
The core argument is about speed. Attackers using AI can find and exploit software flaws faster than defenders can respond. The senators put it plainly: “AI is rapidly increasing the ability to discover and exploit weaknesses in software, and our defenses are not keeping pace,” as CBS12 reported.
The senators want three things from the administration. They ask the White House to speed up development and deployment of AI-powered defensive tools. They ask for security standards for autonomous agents. They also ask that the National Institute of Standards and Technology, known as NIST, write rules for defensive AI agents specifically.
The letter also asks Clayton’s task force to bring AI companies and federal agencies together to secure digital infrastructure. That request treats the problem as shared. No single agency or company can patch the gap alone, and the senators are asking the executive branch to organize the response.
Why banks say they cannot tell rogue agents from authorized ones
The letter’s most concrete claim involves banks. The senators cite reports that major U.S. banks have found unauthorized AI agents accessing customer accounts. Some of those banks, according to the reports, cannot tell authorized agents from rogue ones. The sources reviewed do not name the banks or say how many accounts were affected, so the scale of the problem is unknown.
That gap is the real issue. A bank that cannot classify the software touching an account has trouble deciding what to block. Block too little and criminals or runaway tools get through. Block too much and customers lose access to agents they chose on purpose.
The liability question is already moving. USA 24 has reported that five US banks now make customers liable for losses caused by outside AI agents, which means the risk is showing up in account terms as well as in security reports. Readers with agent-enabled apps should read that bank liability story before assuming a bank will cover an unauthorized transaction.
Japan attacks and the Microsoft outage in the letter
The letter also points to a wave of attacks on Japanese companies and financial institutions. Those attacks caused large-scale data breaches and leaks, and the Japanese government is trying to contain them. The sources reviewed give no dollar figures or record counts, so the size of the damage remains unclear.
Domestic disruptions appear in the argument too. The senators point to a Microsoft outage that affected many services, including email, as one of the recent disruptions behind the letter, according to Fox 4 News. USA 24 has covered the November outages and the fight over their cause, and officials in Washington have not agreed on one explanation.
This letter is not the senators’ first. It follows an earlier October 13 letter from the same two senators urging a national AI safety and cyber defense strategy. No White House response to either letter appeared in the sources reviewed, which is why the new request uses the word “immediately.”
AI agent security standards the senators want from NIST
The sharpest request targets defensive tools themselves. The senators ask NIST to write security standards for defensive AI agents so those agents do not add vulnerabilities or exceed their access limits. That framing treats a security agent as a possible attack surface. A tool that patches systems or monitors traffic usually needs broad credentials, and those credentials become a target if the tool is compromised.
A federal standard of that kind would likely define how much access an agent may hold, how its actions are logged and when it must stop. Businesses and IT teams would feel those rules first, because they would decide which agents may touch systems and money. A separate federal framework for frontier AI labs is also taking shape, as USA 24 has described in its SAFE AI Act outline.
Lawmakers in the House are circling a similar idea. The bipartisan Stop Rogue AI Act, introduced by Reps. Mike Lawler (R-NY) and Josh Gottheimer (D-NJ), would direct NIST to set standards to track AI agents and let humans grant or revoke their access, according to Newsweek. The Senate letter and the House bill point at the same core idea: agents need identities, limits and a way to shut them off.
Democrats’ AI regulation plans for 2027
Democrats prepare to run Congress in January 2027 with AI oversight as a stated priority, as NPR reported. The 120th Congress begins that month. The Trump administration has taken a hands-off approach to AI, rolled back Biden-era AI policies and sought to limit state AI laws. That sets up a clash with the incoming majority.
The push is already taking shape in the Senate. Sen. John Hickenlooper said the next Senate Democratic leader should have a “day one plan” for AI regulation, and a senator has proposed a select Senate committee on AI, according to the Associated Press.
The proposal has drawn resistance. Sen. Ted Cruz, the Republican chairman of the Senate Commerce Committee, opposed taking AI oversight away from his committee. He called one version of the idea “makes no sense,” and said bipartisan talks are already underway for next year.
The veto problem is real. Coverage of the incoming Congress notes that any AI bill passed would face a presidential veto threat. That means any federal agent standard would need to survive both a divided debate and a possible veto fight before it reached law.
How to protect your bank account from AI agents
The letter does not give consumers steps to take, so the following precautions are general good practice rather than recommendations from the senators:
- Review bank alerts and statements for transactions you did not approve, and report them to your bank quickly.
- Limit which apps and AI agents can access financial accounts, and revoke access for tools you no longer use.
- Turn on multi-factor authentication for banking, brokerage and payment accounts.
- Read your bank’s terms to see whether losses from outside AI agent transactions would fall on you.
Businesses should go a step further. Audit which AI agents hold credentials to payment systems, and ask vendors how they log and limit agent actions. Those questions are likely to become standard once federal guidance arrives.
What to watch next
Three questions will shape what happens next. First, whether the White House or Clayton responds to the letter; no reply appeared in the sources reviewed. Second, which banks reported unauthorized agent activity, and whether regulators name them or publish account counts. Third, whether NIST receives a formal assignment to write agent standards, and whether any such standard can clear a likely veto fight.
Some reports describe AI agents escaping a controlled test environment and reaching an outside system. Those accounts come from secondary or low-reliability sources that could not be verified, so this article does not rely on them. The Senate leadership race and the proposed AI select committee also remain unresolved.
The letter sets the agenda for the next Congress. Banks, regulators and lawmakers will have to decide who controls AI agents that can move money and change systems, and the answer will shape how much risk customers carry in the meantime.




