Skip to content
Theme
OnDevice Tools

GadgetsSmart Home

AI agent: major US banks treat all actions as customer‑authorised, even mistakes

Major US banks including Citibank, PNC and Citizens have clarified that any action by a third‑party AI agent is treated as authorised by the customer, even if the agent makes a mistake.

Illustration of a bank customer using an AI assistant on a smartphone, with a warning notice about liability

Key takeaways

  • Banks now treat all AI agent actions as customer‑authorised, even mistakes, shifting liability to users.
  • Citibank, PNC and Citizens have mirrored the policy, refusing responsibility for third‑party AI actions.
  • Google Gemini and Anthropic have introduced read‑only tools to limit agent access to financial data.
In this article
  1. Bank policy change
  2. AI agent access and risk
  3. Industry response
  4. What it means for users

Bank policy change

Illustration: banking, AI, Google logo, Anthropic logo and the labels Anthropic, Citibank, Google

AI agent banking authorization: major US banks including Citibank, PNC and Citizens have clarified that any action by a third‑party AI agent is treated as authorised by the customer, even if the agent makes a mistake. The policy, announced on 20 November 2026, shifts liability onto users who grant the agents access to their accounts. According to TechRadar, the banks say customers are responsible for activities conducted through the agents.

AI agent access and risk

When customers give an AI agent permission to view or move money, the bank considers the customer to have authorised the action. The risk is heightened because mistakes or malicious actions can occur without the user’s immediate knowledge. Some AI providers have responded by introducing read‑only modes; Gemini for Google Wallet and Anthropic both offer tools that let users analyse spending without granting purchasing power.

Industry response

Citibank’s rivals Citizens and PNC have mirrored the stance, refusing to accept responsibility for third‑party AI actions. Other banks are watching closely, as the policy could set a precedent for how financial institutions handle emerging AI services. AI companies are adjusting access levels; Anthropic launched a similar read‑only tool earlier in November, and Google Gemini now limits agent capabilities to analysis only.

What it means for users

Users should organise and review any AI agent permissions they have granted and consider limiting access to read‑only where possible. Banks advise customers to treat any AI‑driven transaction as their own responsibility, especially when credentials are shared. Keeping software updated and using reputable AI services can reduce the chance of unintended actions.

Sources

  1. Major US banks say all AI agent actions are user authorized, even mistakesTechRadar

Tags

Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.