Citi has updated its online banking terms to make customers responsible for the actions of third-party AI agents accessing their accounts, including when an agent leaks credentials or ignores instructions. Under the new terms, actions taken using credentials supplied to an outside agent are treated as authorized by the customer, even if the tool exceeds permissions or shares those credentials. A Citi spokesperson said the November revision clarifies the risks of sharing banking credentials, and that the provision does not apply to AI agents offered by Citi itself.
Other major banks are tightening rules in similar ways. JPMorgan and Bank of America announced comparable policies in October, while Citizens Bank warned customers by email that they could be responsible for what their AI assistants do. PNC changed its digital agreement on Wednesday, distinguishing bank-approved connections from handing an agent unrestricted account access. The moves follow an October 22 banking security scare in which major banks detected AI agents logging into customer accounts, sometimes taking actions that appeared to go beyond user instructions. The Office of the Comptroller of the Currency has since issued nonbinding AI examiner guidance covering outside agents, payment controls, and human approval.
Federal consumer protections still restrict banks from holding consumers liable for genuinely unauthorized electronic transfers, regardless of contractual wording. The Consumer Financial Protection Bureau says banks cannot use agreements to waive those legal protections, though application to customer-authorized AI agents remains unclear. PNC's stricter approach prohibits customers from giving third-party AI agents passwords or one-time security codes, and agents generally cannot initiate payments unless PNC expressly permits it.
Meanwhile, Sumsub's Identity Fraud Report 2026–2027, released on 18 November, found that about 1 in 40 approved users in Asia Pacific (2.4%) is linked to a detected fraud network, and roughly a third of those links appear only after the user has passed identity verification. APAC accounted for 33% of all fraud networks Sumsub detected worldwide. The report lands a day after East Asia Summit leaders met in Manila and agreed to step up cooperation against online scams. The UN Office on Drugs and Crime estimates Southeast Asia-based scam groups caused US$88.3–114.1 billion in losses in 2025, and Chainalysis puts crypto scam revenue for the year at about US$17 billion.
High-profile cases illustrate the pattern. In October 2025, US prosecutors charged Prince Group chairman Chen Zhi over forced-labour scam compounds in Cambodia and moved to forfeit about 127,271 bitcoin, then worth roughly US$15 billion. Cambodia arrested Chen and sent him to China in January. In Hong Kong, the JPEX case, with more than 2,700 victims and HK$1.6 billion in losses, has widened into money laundering with over 80 arrests.
For crypto, the post-KYC gap is a pointed finding for exchanges and custodians. Sumsub's global crypto data shows the fraud rate edged down from 2.2% to 2.1%, and linked or repeated fraud caught at verification fell 7%, but the Fraud Complexity Index for crypto rose 16%, digitally edited IDs jumped 88%, and identity mismatches rose 19%. KYC bypass kits are openly advertised on Telegram targeting Binance, BBVA, and Revolut. In September, Thai police arrested a suspect wanted in China over a face-swap scheme that allegedly took over an OKX account and moved about 4.7 million USDT.
Sumsub's vice president for APAC, Penny Chai, said: "Fraudsters aren't just trying to get past the front door anymore. An account that was legitimate on day one may not be by month six, so exchanges need to keep checking identity against how the account actually behaves over time." She added that crypto holds an advantage because every on-chain transaction leaves a trail, and exchanges responding best combine that traceability with identity data. Sumsub's State of the Crypto Industry 2026 study found 55% of crypto businesses experienced fraud in 2025, and 74% now rank verification accuracy as important, versus 39% for onboarding speed.
Regulators are also focusing on accounts that turn after day one. Thailand had suspended 62,516 mule accounts by 31 July 2026, and Singapore's Scams (Countermeasures) and Other Matters Act 2026 lets police order scam-linked accounts disabled. The wider market still has room to improve: only 25% of APAC businesses surveyed use fraud network analysis. AI agents add a new authorization gap—61% of APAC businesses have detected or suspected fraud involving an AI agent, yet only 15% can suspend or revoke an agent's access. In a crypto example from May, an attacker used a prompt injection to get Grok to issue a transfer command that the Bankr trading bot executed, draining about US$170,000.