Bank of America hires 1,000 human workers to fix AI-detected software flaws

Bank of America has hired 1,000 human contractors to fix software vulnerabilities uncovered by advanced artificial intelligence, offering a striking example of how AI can create new work for humans rather than simply replace them. The hiring followed the discovery of security weaknesses by frontier AI models, including Anthropic’s Mythos, which can identify software flaws that traditional security tools have missed.

The unexpected hiring spree comes as banks and technology companies pour billions into AI to automate tasks, cut costs, and operate with fewer employees. Yet at Bank of America, AI’s ability to uncover hidden problems has triggered an expensive effort requiring thousands of human workers.

Bank of America CEO Brian Moynihan disclosed the hiring during an interview with CNBC on Wednesday, following the bank’s third-quarter earnings release.

“We put on 1,000 contractors to help us do the patching work in the second quarter, and those people remain here today, helping us, on top of the 3,000 people we have in that area,” Moynihan told CNBC.

The contractors are working alongside roughly 3,000 internal staff to repair vulnerabilities across the bank’s proprietary software and code supplied by outside vendors.

The effort has come at a substantial cost. Moynihan said software remediation and regulatory work tied to anti-money laundering compliance added a “couple hundred million plus” to third-quarter expenses. The bank expects those costs to decline as the repairs progress.

Bank of America hired 1,000 contractors to patch software flaws exposed by Anthropic’s Mythos

The hiring follows the introduction of Anthropic’s Mythos model in April through Project Glasswing, an initiative that gave a limited group of banks, technology companies, and government agencies access to advanced AI capabilities for identifying software vulnerabilities.

The technology presents a new challenge for large financial institutions. AI systems can uncover weaknesses buried deep inside software that has run for years, potentially exposing security gaps that conventional testing methods missed.

The AI irony: Bank of America spends hundreds of millions fixing software flaws exposed by Anthropic’s AI

For banks, those discoveries can trigger extensive remediation work. Fixing vulnerabilities in financial systems requires more than identifying defective code. Engineers must assess potential consequences, modify applications, test changes, and confirm that repairs do not disrupt critical banking operations.

That helps explain why AI-driven vulnerability detection can generate substantial demand for human expertise.

The comparison with the Y2K era is hard to miss. Banks once deployed large teams of programmers to inspect and repair aging systems ahead of the millennium date change. This time, advanced AI is helping expose the problems, but people remain deeply involved in resolving them.

The distinction matters. Bank of America’s hiring does not establish that Mythos is incapable of repairing the vulnerabilities it discovers. It shows that detecting security weaknesses and safely deploying fixes across a major financial institution are separate challenges.

Bank of America expects $700 million in AI savings even as security costs rise

The cybersecurity spending comes alongside a broader effort by Bank of America to use AI to reduce operating expenses.

Moynihan said the bank has approximately 130 internal AI projects, with projected annual savings rising from $300 million in 2026 to $700 million in 2027, according to CNBC.

The bank’s workforce has shrunk by roughly 3,400 employees over the past year, and management positions have been reduced by about 5%. The company has not established how much of that reduction it can attribute directly to AI.

That creates an unusual picture of AI adoption inside one of America’s largest banks. The same technology promising hundreds of millions in savings is driving an urgent security response that requires additional workers and significant spending.

Moynihan expects the remediation costs to ease as the bank works through its software vulnerabilities.

“Once you patch, you don’t have to patch as much. … The pipeline moves through,” he said.

The bigger story extends beyond Bank of America. As frontier AI models become better at finding vulnerabilities that older tools missed, companies may discover that their existing software contains far more security weaknesses than previously recognized.

That could create a new challenge for corporate technology budgets: AI may make finding problems cheaper and faster, but fixing them safely can still be expensive.

For all the talk about AI replacing programmers, Bank of America’s experience offers a different lesson. AI can dramatically increase what organizations can discover without eliminating the human expertise needed to act on those discoveries.