Bank of America staffs AI flaw fixes amid AML fines

Bank of America has about 4,000 people fixing AI-found flaws, with disputed hundreds of millions in costs and large AML penalties.

Bank of America CEO Brian Moynihan said the bank brought in about 1,000 contractors in the second quarter, still on site, to work with roughly 3,000 technical employees on software vulnerabilities found with AI. The Register described that work as patching Anthropic Mythos findings, The Decoder said Mythos and other models, and CSO Online said the flaws surfaced after an early Claude Mythos Preview that Anthropic told a House committee had uncovered thousands of high-severity issues. Cost accounts conflict: the Register cited a couple of hundred million dollars or more in combined Mythos and anti-money-laundering spending without a split, The Decoder said a few hundred million dollars in the third quarter for software work and AML together, and CSO said a couple of hundred million on third-quarter AML remediation plus a similar separate amount on vulnerability fixes, along with billions spent on technology. Only the Register itemized a $1.21 billion FinCEN settlement over unreported suspicious activity and national-security failures, a $750 million OCC penalty, and an $11.6 million OFAC settlement; CSO described a $1.2 billion money-laundering fine. The Register also reported a 17 percent profit drop and a share-price decline of more than 8 percent, with its bullets placing those moves in the third quarter, and separately said Mozilla reported 271 Firefox 150 flaws found by Mythos. The Decoder added that Anthropic released Mythos in April only to Project Glasswing partners after finding thousands of serious flaws, that OpenAI launched Daybreak in May, and that Moynihan expects remediation costs to decline.

• CEO Brian Moynihan said Bank of America hired about 1,000 contractors in Q2, still on site, alongside roughly 3,000 technical employees—about 4,000 people—to fix AI-found software flaws; CSO cited about 1,000 consultants for vulnerability…
• Sources differ on the AI source: The Register said Anthropic Mythos findings; The Decoder said Mythos and other models; CSO said an early Claude Mythos Preview.
• Spending figures conflict: a couple of hundred million dollars or more combined (Register) or a few hundred million in Q3 combined (Decoder), versus CSO's couple of hundred million on Q3 AML remediation plus a similar separate amount on…
• The Register itemized a $1.21 billion FinCEN settlement, a $750 million OCC penalty, and an $11.6 million OFAC settlement; CSO described a $1.2 billion money-laundering fine.
• The Register reported profit down 17 percent and shares down more than 8 percent, with bullets placing those moves in Q3 while the narrative ties settlements to the same quarter as Q2 comments.
• Anthropic released Mythos in April only to Project Glasswing partners after finding thousands of serious flaws and told a House committee it found thousands of high-severity flaws; Mozilla separately said Mythos found 271 Firefox 150 flaws.
• OpenAI launched Daybreak in May, and Moynihan expects remediation costs to decline.