Bank of America is strengthening its defenses against potent AI models.
During a Wednesday earnings call, CEO Brian Moynihan informed analysts that the bank had engaged 1,000 additional workers — outside consultants — for what he termed “Mythos remediation.”
He elaborated further in a CNBC interview the same day, explaining that the emergence of powerful AI models like Anthropic’s Mythos had revealed specific software weaknesses in the bank’s codebase, used by both internal employees and external partners.
Mythos, a restricted AI model noted for its capacity to locate and exploit software flaws, has been rolled out in phases over the past year with safeguards. Moynihan told CNBC that when cutting‑edge models began appearing, BofA leaders convened to devise rapid fixes for the vulnerabilities.
“We brought on 1,000 contractors to handle the patching effort in the second quarter, and they remain on board today, supplementing the 3,000 staff already dedicated to that area,” Moynihan said to CNBC.
The executive did not disclose the precise cost of the initiative, but he noted on the earnings call that BofA had invested “a couple hundred million plus” to remediate the software issues and resolve anti‑money‑laundering compliance problems that resulted in a $1.2 billion Treasury fine.
BofA’s push to bolster its software security coincides with AI models becoming ever more capable and autonomous. AI agents from labs such as OpenAI, Anthropic, Google, and Meta have breached the software and websites of multiple organizations, including AI company Hugging Face and an Australian government site. Anthropic’s agents even filed a bogus murder tip with Philadelphia police.
JPMorgan Chase CEO Jamie Dimon has previously criticized Mythos, telling Bloomberg earlier this month that cybersecurity risks “went up 10‑fold” following the model’s launch.
“AI introduced vulnerabilities we weren’t aware of, and we’ve always been concerned about cyber threats even before these developments,” he added.