The newest source material points in the same direction from different angles: firms are adopting AI tools, but the real operational question is whether the workflow is secure, scoped, and useful enough to change how work gets done. Password reuse can expose the systems that AI touches, agentic attacks show how AI can be used against financial organizations, and legal AI may still leave firm operations largely unchanged when the incentive structure stays the same.
Why AI workflow design starts with identity and access
A firm cannot safely automate around weak account security. The password-reuse warning is especially relevant for firms that connect tax software, client portals, email, and bank access, because one reused credential can expose multiple systems.
For custom AI and agentic workflows, that means identity controls come first: unique logins, multifactor authentication, password manager use, and breached-password screening before any workflow is allowed to touch sensitive client data or financial systems.
What agentic AI changes for law and accounting firms
The reporting on AI-led hacks shows that attackers are already using AI tools to probe systems and steal data from financial organizations. That matters for professional-services firms because the same kinds of automation that improve productivity can also accelerate malicious testing and compromise.
Firms building internal AI assistants, document workflows, or client-service agents should assume the workflow will be part of the security surface. Access should be limited to the smallest practical set of systems and actions, especially where client information, tax data, or banking functions are involved.
Why many firms see AI but not transformation
The legal AI commentary points to a familiar problem: firms can buy tools without changing how work is actually organized. If the workflow still relies on the same bottlenecks, approvals, and risk fears, the technology may save time on the margins but fail to move the operating model.
For law firms and accounting firms, custom AI is most valuable when it is tied to a specific process with a clear owner, a narrow handoff, and a measurable business problem. Broad promises usually produce broad disappointment.
How to scope a practical custom AI or automation project
The state e-file coverage piece is a useful reminder that even one apparently simple workflow can hide a lot of complexity. In the same way, an AI project should be scoped around the exact task, exact systems, and exact approval path-not around a vague promise that the tool supports everything.
A better starting point is a workflow with repeatable inputs, low ambiguity, and a clear error check. That is where custom AI, automation, or an agentic assistant can reduce busywork without creating a new risk problem.
- Treat identity and access as part of the AI workflow, not as a separate IT issue.
- Start with narrow, repeatable processes instead of trying to automate the whole firm at once.
- Assume AI adoption will not change operations unless incentives, controls, and handoffs change too.
- Use vendor and workflow due diligence to test what a tool can actually do in your environment.
Sources watched
- How to Get Your Firm Breached: Use the Same Password Everywhere (CPA Practice Advisor AI)
- Legal AI vs The Adversarial Incentive (Artificial Lawyer)
- Claude Lies to Police, AIs Nobody Wants, & AI-led Hacks (The White Box)
