More banks warn AI agent users are on their own, even if hacked
agents
| Source: TechCrunch | Original article
Citi joins other banks in warning that users of personal‑finance AI agents bear full responsibility for security, even if the service is compromised.
Citi has joined a growing list of U.S. banks that are warning customers they are effectively on their own when they grant personal‑finance AI agents access to bank or brokerage accounts. The move follows recent disclosures that major financial institutions are detecting third‑party AI tools logging into customer accounts without any bank‑issued authorization. In many cases the agents operate independently of the banks’ security controls, leaving the institutions unable to verify what actions the software is permitted to take or to intervene if the agents behave maliciously.
The development matters because it highlights a new frontier of trust risk. Consumers are already hesitant to share email or chat histories with AI services, and extending that reluctance to financial data creates a “trust challenge,” analysts note. Deloitte’s recent survey found that 83 % of banking customers would feel anxious about an AI agent acting on their finances without explicit approval, and two‑thirds believe the potential for error outweighs any benefit. If an AI‑driven tool is compromised, the bank’s liability is limited; the onus falls on the user to protect their credentials and monitor activity.
What to watch next is how banks will respond to the gap between consumer demand for convenient AI‑powered budgeting, investing and payment services and the lack of regulatory or technical safeguards. Industry observers expect tighter disclosure requirements, possible restrictions on third‑party integrations, and the emergence of bank‑issued “agent‑friendly” APIs that could give institutions visibility into, and control over, external AI actions. The next wave of policy and product decisions will determine whether AI agents become a trusted part of personal finance or remain a risky, user‑borne experiment.
Sources
Back to AIPULSEN