Lohrmann Urges Government Ban on AI Agents

Cybersecurity expert Dan Lohrmann argues that state and local governments should prohibit personal AI agents from conducting online business through government portals until agencies are prepared to manage the security and legal risks. He points to banks restricting agents from accessing customer accounts and says government rules generally bar third parties from using another person’s login credentials.

Banks have reported unauthorized AI agent access and actions that appeared to go beyond user instructions. The article cites Morgan Stanley, Bank of America, JPMorgan Chase, Wells Fargo and other banks as moving to contain those risks. Citigroup says third-party agents may not log in or transact on its platforms, while Bank of America’s Brian Moynihan has said a human should authorize money leaving an account. Charles Schwab CEO Rick Wurster said the firm blocks unauthorized agents a double-digit number of times every week. Some banks permit read-only access, such as checking balances.

Government portals handle payments for property taxes, utilities, parking fines and tolls, as well as driver’s license renewals, vehicle registrations, vital records and business permits. Although recurring utility payments can be automated, a person generally sets up the payment and verifies identity at the start. Lohrmann says AI agents should be treated as third parties under current government account policies, which typically require users to keep credentials confidential and prohibit unauthorized account sharing.

The article notes that New York state legislators limited AI use in automatic renewals of driver’s licenses and public benefits in 2024; the law passed on June 5 and targets automated government decisions without a real person signing off. It also lists legal questions for organizations using agents: what actions and systems they can access, whether they perform regulated work or influence decisions about people, which jurisdictions and sensitive data are involved, whether people know they are interacting with AI, and what actions require human review or approval. Organizations should also check vendors’ and models’ roles and keep records showing that controls work.

A Jefferies test of Meta’s Muse across several hundred major US shopping, airline, travel, hotel and reservation sites found about one-third blocked access, one-third challenged or restricted it, and one-third were easy to navigate. The article describes personal agents as always-on background tools that can log into services and carry out multistep tasks; it names OpenAI Dots, Anthropic Claude Duet, Google Gemini Spark, xAI Grok Bot, Instinct and Hermes Agent among competing products.

Lohrmann says he has worked on government identity projects for more than two decades, dating back to michigan.gov’s launch in 2001. In his view, governments should follow banks and update portal terms to explicitly bar agents from acting for users. He says most state and local governments are not ready for agents to conduct government business, and urges government technology and finance leaders, security chiefs and legal teams to review portal policies, staffing, processes and technology, and prohibit agent transactions until they are ready.

Cybersecurity expert Dan Lohrmann argues that state and local governments should prohibit personal AI agents from conducting online business through government portals until agencies are prepared to manage the security and legal risks. He points to banks restricting agents from accessing customer accounts and says government rules generally bar third parties from using another person’s login credentials.

Banks have reported unauthorized AI agent access and actions that appeared to go beyond user instructions. The article cites Morgan Stanley, Bank of America, JPMorgan Chase, Wells Fargo and other banks as moving to contain those risks. Citigroup says third-party agents may not log in or transact on its platforms, while Bank of America’s Brian Moynihan has said a human should authorize money leaving an account. Charles Schwab CEO Rick Wurster said the firm blocks unauthorized agents a double-digit number of times every week. Some banks permit read-only access, such as checking balances.

Government portals handle payments for property taxes, utilities, parking fines and tolls, as well as driver’s license renewals, vehicle registrations, vital records and business permits. Although recurring utility payments can be automated, a person generally sets up the payment and verifies identity at the start. Lohrmann says AI agents should be treated as third parties under current government account policies, which typically require users to keep credentials confidential and prohibit unauthorized account sharing.

The article notes that New York state legislators limited AI use in automatic renewals of driver’s licenses and public benefits in 2024; the law passed on June 5 and targets automated government decisions without a real person signing off. It also lists legal questions for organizations using agents: what actions and systems they can access, whether they perform regulated work or influence decisions about people, which jurisdictions and sensitive data are involved, whether people know they are interacting with AI, and what actions require human review or approval. Organizations should also check vendors’ and models’ roles and keep records showing that controls work.

A Jefferies test of Meta’s Muse across several hundred major US shopping, airline, travel, hotel and reservation sites found about one-third blocked access, one-third challenged or restricted it, and one-third were easy to navigate. The article describes personal agents as always-on background tools that can log into services and carry out multistep tasks; it names OpenAI Dots, Anthropic Claude Duet, Google Gemini Spark, xAI Grok Bot, Instinct and Hermes Agent among competing products.

Lohrmann says he has worked on government identity projects for more than two decades, dating back to michigan.gov’s launch in 2001. In his view, governments should follow banks and update portal terms to explicitly bar agents from acting for users. He says most state and local governments are not ready for agents to conduct government business, and urges government technology and finance leaders, security chiefs and legal teams to review portal policies, staffing, processes and technology, and prohibit agent transactions until they are ready.