Bank of America says machine learning tools it deployed to catch financial crime were not enough, after regulators handed the lender a $1.2B penalty over weak anti-money-laundering controls.
Chief executive Brian Moynihan told analysts on the third-quarter call that the bank has spent billions on technology and assigned thousands of staff to financial crime work, with a couple hundred million more going to remediation in the quarter alone.
Cleaning up flawed software took a separate pile of money and an unusual step: hiring 1,000 contractors to work through vulnerabilities the bank learned about after receiving an early copy of Anthropic’s Claude Mythos Preview. The Register reported last week that the same model drove the contractor surge.
A test case for AI-driven patching
Anthropic told the House Financial Services Committee that its preview build had already surfaced thousands of high-severity flaws, including some in every major operating system and browser. For a bank, that volume turns into a staffing problem rather than a research curiosity.
Security leaders should read the episode two ways. Model-assisted code review finds real bugs fast, and it also manufactures remediation debt faster than change boards can absorb. Budget for the fixers, not just the finder. Firms that skip the second half will find regulators asking why known defects sat open.
