To patch software flaws brought to light by advanced AI models, Bank of America hired 1,000 outside contractors, CEO Brian Moynihan told CNBC — staff added on top of the 3,000 people already assigned to that area. Remediation and associated compliance work ran the bank “a couple hundred million plus” during the third quarter.
Few major US corporations have attached hard numbers to the defensive bill run up by frontier AI, and this ranks among the first such disclosures. Anthropic’s Mythos model, which reached a limited set of banks, technology firms and government agencies in April, is built to uncover and weaponize software flaws that went unnoticed for decades.
Both sides of the same coin
Calling the risk genuine, Moynihan said: “That risk is real risk, and that’s why when the frontier-type models came out, a lot of us sat down and said, how do we figure out how to get through this quickly?”
He cast the outlay as a passing phase. New findings should taper off as vulnerable code gets fixed. For the time being, the contractors stay.
Elsewhere in the business the same technology cuts the other way. Some 130 internal AI projects are underway at Bank of America, and Moynihan told analysts their savings should rise from $300M this year to $700M in 2027.
An industry-wide bill
Enterprises elsewhere now have a benchmark that most have declined to publish, courtesy of Moynihan’s figures. Shares of the bank slid as much as 8 percent on the day he spoke, a sign that investors were uneasy about near-term costs rather than skeptical of the risk itself. Should models like Mythos keep surfacing what human auditors missed for decades, remediation budgets in financial services will read less like a one-off cleanup and more like a standing expense.
