Bank of America has brought in 1,000 external contractors to help patch software vulnerabilities identified through work involving advanced AI models, including Anthropic's Mythos. The move highlights the growing cybersecurity demands facing financial institutions as AI becomes more capable of finding weaknesses in complex software systems.

On 14 October 2026, CEO Brian Moynihan told CNBC that the bank had hired 1,000 external contractors during the second quarter to help patch vulnerabilities across its own software and systems supplied by third-party vendors. The contractors remain in place, working alongside approximately 3,000 existing staff in the relevant area.

The initiative reflects a wider challenge for banks. AI tools can help organizations identify and repair software weaknesses, but similar capabilities could also make it easier for attackers to discover and exploit vulnerabilities. For financial institutions, the issue is no longer simply whether they have cybersecurity teams, but whether those teams can identify, prioritize, and remediate weaknesses quickly enough as AI models keep improving.

The financial implications are becoming clearer too. Moynihan said Bank of America spent "a couple hundred million plus" during the third quarter on software remediation and anti-money-laundering compliance issues, and he expects remediation costs to decline as the backlog gets addressed. At the same time, the bank expects its internal AI projects to generate significantly greater savings, illustrating the technology's two-sided financial impact.

Why Bank of America Added 1,000 Contractors

Moynihan said Bank of America mobilized thousands of employees and outside personnel earlier in 2026 to strengthen its internal software and vendor-supplied code. The effort accelerated as frontier AI models demonstrated stronger capabilities in software analysis and cybersecurity tasks. The bank brought in 1,000 contractors during the second quarter to help complete the patching work, and they remain alongside roughly 3,000 staff already working in the area.

The bank did not disclose a standalone cost for the Mythos-related remediation effort. Moynihan's third-quarter spending estimate also included anti-money-laundering compliance issues, so the combined figure should not be read as the cost of AI-related cybersecurity work alone.

Large banks depend on extensive technology environments, including internally developed applications, legacy systems, and products supplied by external vendors. A weakness in one component can create risks elsewhere if systems are connected or share access to sensitive information. Patching involves more than finding a flaw. Security teams must assess severity, determine which systems are affected, test a correction, deploy it safely, and verify the fix without disrupting critical services. Adding external specialists increases available capacity, though the response's effectiveness still depends on remediation quality and ongoing monitoring.

What Makes Anthropic's Mythos Significant?

Anthropic introduced Claude Mythos Preview in April 2026 as a general-purpose model with particularly strong capabilities in coding and computer security tasks. The company launched Project Glasswing to help organizations responsible for critical software use these capabilities defensively. Anthropic said Mythos Preview had identified thousands of high-severity vulnerabilities, including weaknesses in major operating systems and web browsers, and the model is available through a controlled research preview rather than an unrestricted public release.

The underlying concern is that models capable of understanding complex code may help discover flaws that traditional testing has missed. If those capabilities become more widely accessible, defenders could benefit from faster discovery, but malicious actors could also use AI to accelerate exploit development. For banks, this changes the pace of the challenge, since a vulnerability undiscovered for years may suddenly surface through new analysis techniques and must then be fixed across every system in which it appears.

Identifying a vulnerability, though, does not automatically mean a system has been breached. The distinction between discovering a weakness and carrying out a successful attack remains important. Bank of America's reported response concerns software remediation, and the available reporting does not establish that Mythos successfully breached the bank or that customer data was compromised.

Project Glasswing and the Push for Defensive AI

Project Glasswing brings together Anthropic and major technology and infrastructure organizations to use advanced AI capabilities to secure critical software. Launch partners included Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorgan Chase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks. The program aims to help participants identify and fix vulnerabilities in foundational systems while sharing lessons that strengthen security across the wider ecosystem.

This defensive approach reflects a broader shift in cybersecurity, where AI is increasingly viewed as a means of testing software and supporting security research, not just a productivity tool. Still, AI-assisted security is not a substitute for established controls. Banks still need secure development practices, access management, monitoring, and rigorous testing before changes reach production. The challenge is using AI to widen security analysis while maintaining human accountability.

The Cost of Cybersecurity Remediation

Moynihan's comments also brought attention to the financial burden of responding to software vulnerabilities. He told CNBC that Bank of America spent "a couple hundred million plus" during the third quarter on software remediation and anti-money-laundering compliance issues, and he expects those costs to decline as the pipeline of outstanding fixes shrinks.

A concentrated effort to address a backlog can require temporary staffing, and once the most urgent issues are resolved, the need for extra personnel may ease. Banks cannot treat cybersecurity as a one-time project, though, since new vulnerabilities emerge as software changes and attackers develop new techniques. The combined cost Moynihan reported also shows how multiple technology and regulatory priorities can hit a bank's expenses at the same time.

AI Savings Offer a Different Side of the Equation

While Bank of America is increasing its defensive cybersecurity effort, the bank also expects its own AI initiatives to generate greater efficiencies. Moynihan said the bank has approximately 130 internal AI projects, with savings expected to rise from $300 million in 2026 to $700 million in 2027.

These figures represent management expectations, not guaranteed results, and the bank has not said all projected savings are tied to cybersecurity. Still, the comparison highlights the different ways AI affects a financial institution's economics. AI tools can help employees work more efficiently, but advanced models can also expose weaknesses that institutions must then repair, creating additional short-term costs. The net impact depends on how effectively a bank deploys AI and controls the associated risks.

What This Means for the Banking Sector

Bank of America's experience offers several lessons. Vulnerability management must keep pace with AI, since banks need processes that handle a growing volume of findings without letting critical issues go unresolved, prioritized by severity and exposure to external threats.

Third-party software matters too. The bank's reported effort covered both proprietary and vendor-supplied code, so institutions need visibility into the software they depend on and reliable processes for validating security updates. Additional staffing can help address urgent backlogs, but long-term resilience depends on sustainable engineering practices and clear accountability.

AI defense needs governance as well. AI can support security teams, but organizations still need to validate findings and assess proposed fixes before deployment, since automated analysis should complement rather than replace established security processes. Cybersecurity remains a financial and operational priority, since weaknesses at a large bank can affect service continuity, customer trust, and regulatory compliance.

Bank of America's Wider Financial Picture

The cybersecurity comments came as Bank of America reported third-quarter 2026 results: net income of $6.9 billion, or $0.91 per diluted share, including a $1.2 billion regulatory settlement charge that reduced diluted earnings per share by $0.17. Excluding that charge, adjusted net income reached $8.1 billion, with adjusted diluted earnings per share of $1.08, non-GAAP measures distinct from the bank's reported results.

The quarter's results provide useful context for Moynihan's comments on expenses, since software remediation and compliance spending sat within a broader environment where the bank managed technology risk, regulatory obligations, and AI investment simultaneously. The bank also reported $30.0 billion in revenue, net of interest expense, for the quarter, and the remediation and compliance spending should not be treated as a separately disclosed Mythos-only expense.

The Bottom Line

Bank of America's decision to add 1,000 contractors shows how seriously major financial institutions are treating the security implications of advanced AI. The immediate task is patching weaknesses across internal and vendor software. The longer-term challenge is maintaining secure systems as AI makes software analysis faster and more powerful.

For banks, AI is becoming both a productivity opportunity and a source of additional security pressure. The institutions best positioned to benefit will pair innovation with disciplined vulnerability management, strong vendor oversight, and reliable operational controls, since AI-accelerated software discovery means financial institutions must accelerate the work required to keep their systems secure too.