Banks' AI gains and AI risk are now the same story

Efficiency gains get a body count

DNB said it is reducing its Technology & Services workforce by about 400 full-time equivalents, and it names customer data and Know Your Customer work, along with technology development and coding, as areas where AI agents can already contribute "substantial efficiency gains."

The index says the same

That lines up with what Evident's AI Index found this year: the average bank score across its ranking rose 26%, nearly three times the pace of previous years. BBVA, which the index ranks first among eurozone banks, says software development productivity is up close to 28% and that seventy-five percent of employees already use generative AI on a recurring basis.

More data goes to work

Plaid, meanwhile, added new models this week: LendScore 2 uses cash flow underwriting to predict a borrower's ability to repay a loan with 42% greater strength than traditional credit data alone, and a new foundation model purpose-built for fraud is trained on patterns from the Plaid Network.

The same firms are targets

Korea's financial authorities issued a consumer alert and began a monthlong response effort this week after suspected AI-assisted cyberattacks stole personal information from seven financial firms, including Shinhan Bank and Hyundai Capital. Brokerages, insurers and card issuers are stepping up security checks, and regulators and industry officials suspect the attackers were scanning a broad range of companies for weaknesses rather than pursuing a single target.

What to watch next

The lesson for anyone running agents in KYC, coding or credit decisions is that the efficiency case and the exposure case are now the same case: whatever an agent can read or move, something else is trying to read or move too. So before the next productivity number comes in, check whether the same systems have been tested against exactly that.