AI without the right data, purpose, and response process can become a band-aid. Financial crime prevention needs a connected, 360-degree approach.
Bank of America reportedly hired 1,000 additional contractors to address software vulnerabilities identified by advanced AI models, including Anthropic’s Mythos. According to Cybernews, the contractors joined approximately 3,000 people already working in that area. The report does not establish that those vulnerabilities resulted in a successful attack. Read the Cybernews report.
Finding weaknesses before criminals exploit them is valuable. But the scale of the response raises a broader question for financial institutions: when AI reveals more risk, how quickly can the organization act?
A model can identify a problem. Reducing the risk requires connected information, clear priorities, accountable teams and controls that can respond.
That distinction matters just as much in financial crime prevention.
A pattern is emerging.
Bank of America’s story sits alongside a broader move towards AI-assisted defense.
In April, AWS said it was applying Claude Mythos Preview to critical codebases already undergoing continuous AI security review. Its announcement places the model within an established security operation. Read the AWS announcement.
JPMorganChase also joined Project Glasswing. In the initiative’s announcement, its Chief Information Security Officer described an opportunity to evaluate next-generation AI tools for defensive cybersecurity, with a rigorous, independent approach to deciding how to proceed. This was a commitment to evaluate, not a disclosure of findings or remediation costs comparable to Bank of America’s. Read the Project Glasswing announcement.
Taken together, these developments point to a clear direction: increasingly capable AI is becoming part of institutional defense. The operational challenge is making its output useful, manageable and actionable.
For financial crime teams, the lesson is straightforward. Buying intelligence is only one part of building protection.
AI disconnected from data is a sticking plaster.
Consider an AI system assessing a payment with limited transaction information. It might recognise an unusual amount or destination. But can it see that the customer has just changed devices, reset their credentials and added a new beneficiary? Can it identify connections between the recipient and other suspicious accounts?
Without that context, even a capable model works with an incomplete picture.
Adding AI on top of fragmented systems may accelerate individual tasks while leaving the underlying blind spots intact. The result can be more alerts, more manual reconciliation and more pressure on investigators.
AI without a reliable connection to relevant data and decisions risks becoming a sticking plaster rather than a lasting solution.
Connecting data does not mean giving a model unrestricted access. It means making the information needed for a defined task available accurately, promptly and under appropriate controls.
Choose the application. Avoid “spray and pray”
The starting question should be specific: which decision needs to improve?
Prioritising an investigation, identifying linked accounts and assessing a payment before execution are different problems. They require different data, response times and levels of oversight.
Select and evaluate AI against that purpose. Rules, analytics and human judgement also have roles to play.
Success should be measured through outcomes: better detection, fewer unnecessary interventions, faster investigations and reduced losses. Deploying AI widely is not itself evidence of better protection.
What a 360-degree approach changes
A connected financial crime prevention approach brings customer identity, account relationships, transaction activity, behavioural signals and investigation outcomes into a coherent decision process.
Imagine an attacker gains access to a customer’s account. A software security programme must address the route of compromise. Financial crime controls can provide another opportunity to intervene: recognising an unusual session, a newly added beneficiary and an unexpected transfer together, then triggering additional verification or review.
Depending on the circumstances, that intervention could prevent the fraudulent payment or limit the loss.
This is where a 360-degree approach adds value. It reduces the gaps between individual signals and gives teams a clearer basis for action across prevention, detection and investigation.
It does not eliminate every vulnerability or replace cybersecurity. It strengthens the institution’s ability to stop a weakness from becoming a financial loss.
Build the foundation and the strategy for AI to work
The Bank of America story illustrates how valuable discovery can create a substantial response workload. Detection and response capabilities need to develop together.
For NOTO, this is the case for a connected approach to financial crime prevention: relevant data, coordinated controls and AI applied where it improves a defined decision.
Specific financial crime attacks and challenges arise regularly. Each may require a tailored strategy, depending on the available data, the operating environment and the individuals involved. A 360-degree view provides the foundation for understanding those differences and deciding how to respond.
At NOTO, we work with clients to build these strategies and deploy the capabilities needed to put them into practice, helping prevent financial crime where possible and minimize its impact when it occurs.



