Bank of America hired contractors to protect its software from powerful rogue AI agents

Bank of America is shoring up its defenses against mighty AI models.

CEO Brian Moynihan told analysts in a Wednesday earnings call that the financial institution had hired 1,000 additional personnel — external consultants — for what he called "Mythos remediation."

He expanded on that in an discussion alongside CNBC on Wednesday. He stated the betterment of mighty AI models, specified as Anthropic's Mythos, had exposed certain application vulnerabilities inside the bank's code, which is used by the two inner personnel and external vendors.

Mythos, a restricted AI example known for its capability to discover and utilize application weaknesses, has been released in stages complete the former twelvemonth alongside guardrails. Moynihan told CNBC that whenever frontier models started coming out, BofA executives sat downward and brainstormed how to quickly coast up vulnerabilities.

"We put on 1,000 contractors to assistance us do the patching activity in the second quarter, and those group remain current today, assisting us, on top of the 3,000 group we have in that area," Moynihan told CNBC.

The administrator did not say exactly how much the undertaking cost. However, he stated on the earnings call that BofA had spent a "couple hundred myriad plus" to fix the application problems and location anti-money-laundering compliance issues that led to a $1.2 milliard Treasury penalty.

BofA's funding in beefing up its application safety comes as AI models get increasingly capable and autonomous. AI agents by labs akin OpenAI, Anthropic, Google, and Meta have hacked into the application and website of multiple organizations, including AI resolute Hugging Face and Australia's authorities website. Anthropic's agents equal submitted a false murder tip to Philadelphia police.

JPMorgan Chase CEO Jamie Dimon has earlier spoken out against Mythos, telling Bloomberg before this duration that cybersecurity risks "went up 10-fold" following the model's release.

"AI created vulnerabilities that we didn't cognize about, and we continually concerned concerning cyber before these things," he added.