Bank of America Mythos Contractors: 1,000 Workers Outside a Headcount Down 3,473

Bank of America’s Mythos contractors, about 1,000 people hired to repair software flaws found by artificial intelligence (AI), do not appear in the bank’s headcount. Chief executive Brian Moynihan disclosed them on 14 October, the day the bank reported third-quarter results. The same results show 209,911 employees at the end of September, 3,473 fewer than a year earlier. Both numbers trace back to AI and automation, and only one of them is counted as jobs.

The contractors arrived in the second quarter to fix weaknesses exposed by Mythos, Anthropic’s restricted model, and other frontier systems, working alongside roughly 3,000 employees already on the task. The bank engaged them as outside services, so their cost lands in expenses rather than in the employee total. That classification decides what investors, regulators and the public see when they ask how AI is changing work inside a bank.

What Mythos found and why a bank needed 4,000 people to fix it

Anthropic released Mythos in April to a restricted circle of lenders, tech firms and public bodies. The company said in May that the model had turned up more than 10,000 vulnerabilities across the 50 partner organisations that had access at the time. Mozilla alone logged 271 security vulnerabilities in Firefox that Mythos identified in April, a case we examined in our piece on Mythos and Firefox security.

Each flaw a model finds still needs a person to read the code, write a patch, test it and deploy it without disrupting systems that move money. The work covers code written by the bank’s vendors as well as its own. Moynihan explained to CNBC that the volume of findings, and the speed needed to deal with them, pushed the bank to look beyond its own staff, and he called the risk a real one.

Where the 3,473 jobs went

Bank of America’s third-quarter earnings release puts headcount at 209,911, down from 213,384 a year earlier. Chief financial officer Alastair Borthwick attributed the fall to processes being automated. Moynihan told analysts that manager roles fell by about 5% and that client-facing roles grew.

The release also says AI tools now reach over 200,000 staff, across upward of 300 approved use cases. The reductions sat in automated processes and management layers, according to the two executives, and the growth sat with staff who serve clients directly.

Why the repair work stays outside the headcount

The 1,000 contractors were engaged as consultants, and Moynihan said on the third-quarter earnings call that they are not counted in headcount. Their cost appears in expenses, which rose by about $1.2 billion year on year once the regulatory settlement charge is excluded. Moynihan attributed about $450 million of that rise to wealth management incentives and about $300 million to markets, with most of the remainder going to anti-money laundering (AML) work and Mythos remediation.

The bank reports those last two items together, so the share that belongs to Mythos alone is not public. The AML work follows a $1.2 billion regulatory settlement charge booked this quarter under consent orders with US regulators. Moynihan expects the Mythos portion to decline over time. He added that the bank stays aware that the firms writing the software also want to sell it services.

The return Bank of America expects from AI

Moynihan put the benefit of the bank’s AI projects at about $250 million to $300 million this year, rising to about $700 million in 2027. Roughly 130 projects are live, each with its own payback case, and he expects them to return two to three times their cost over time. He also expects the bank’s technology development budget, about $4 billion a year, to produce roughly 10% more output in 2027 for the same spend.

Those figures put AI in the same frame as any other financial technology investment, from fintech tools to in-house software: projects that each have to show a return. The Mythos contractors sit outside that frame, as a cost created by what frontier AI exposed rather than by any project the bank chose.

Counting AI’s effect on jobs with both numbers

Layoff trackers count the cuts companies announce, which is why AI accounts for a small share of them in the real AI layoff number. Work that AI adds through contracts never enters those tallies, and at Bank of America about 1,000 people are doing exactly that kind of work.

Bank of America’s results give professionals watching AI enter their own organisations a fuller reading than most: fewer managers and process roles, more client-facing staff, and a new category of repair work bought from outside. JPMorgan chief executive Jamie Dimon said on 6 October that his bank is also working to fix vulnerabilities that AI exposed.

Bank of America’s quarter shows AI moving part of a bank’s workforce from the payroll to the procurement budget, where 3,473 fewer employees and 1,000 outside specialists repairing what a frontier model found are reported in different places and rarely read side by side.

Editor’s note

Every piece goes through careful verification, but mistakes can happen. Readers who spot an error or have additional information can write to rosalia@thebrightminded.com.