Bank of America took on 1,000 contract workers in the second quarter to repair software vulnerabilities turned up by advanced AI models, and most of them are still on the payroll.
Chief executive Brian Moynihan disclosed the hiring on CNBC after the bank’s third-quarter results. The contractors sit alongside roughly 3,000 existing staff who handle patching, working across both in-house code and software bought from outside vendors.
The trigger was frontier model capability. Anthropic’s Mythos model, released in April through an initiative that gave a selected group of banks, technology firms and agencies access to vulnerability-hunting AI, flags weaknesses that conventional scanning tools miss.
The bill has not been small. Moynihan said software remediation plus anti-money-laundering regulatory work added a couple hundred million dollars to third-quarter expenses, and expects the cost to ease as repairs land.
The story inverts the usual automation narrative. Rather than removing roles, better detection created a wave of manual work, because finding a flaw is only the first step. Engineers must judge the consequences, rewrite the code, test it and then prove the fix holds.
It is a pattern worth watching for startups selling into regulated industries. AI that widens the funnel of discovered problems also widens the human backlog behind it, and somebody has to pay for that queue.
