The latest news across legal and accounting AI points to a common operating reality for professional-services firms: AI value depends on the workflow around it, and security has to hold up when people, tools, or devices change.
AI adoption is still uneven, and that creates shadow workflow risk
A recent Thomson Reuters survey, as reported by Artificial Lawyer, found that AI adoption among lawyers, accountants, and compliance professionals is often slow and chaotic. It also found that one third of those sampled are using unapproved shadow AI, and that share rises among people who feel their organization is moving too slowly on AI.
For firm leaders, that is not just a technology issue. It is a workflow issue. When people cannot get approved tools that fit their daily work, they route around the system. Custom AI programs and agentic workflows should be designed to reduce that pressure by giving staff approved, secure ways to complete common tasks.
MFA problems show why AI and automation must fit firm operations
CPA Practice Advisor AI highlighted a practical example: when staff change mobile phones or devices, multi-factor authentication can become a real operational hurdle. The article notes that MFA is one of the most important security controls a firm can maintain, and that it is required under the FTC Safeguards Rule for systems that access customer financial information.
That matters for any firm building custom AI or automation. If a client-facing or internal workflow depends on a device-based login or a fragile handoff, the process can break at the wrong time. Secure automation should include account recovery, device transition planning, and clear operational ownership so security does not turn into a bottleneck.
For legal AI, the scaffold matters as much as the model
Artificial Lawyer reported on research showing that the same AI model can perform differently depending on the scaffold around it. The study pointed to context, workflow logic, prompt improvement, planning, agentic loops, retrieval, and tool calling as factors that change the result.
That is a direct lesson for law firms and accounting firms evaluating custom AI. A general model name does not tell you how well the system will handle your work. The real question is whether the workflow is built for the task, whether the model can use the right tools, and whether the output is evaluated against the standards your firm actually needs.
Benchmarks are becoming a decision tool, not a marketing line
Artificial Lawyer also reported on LegalOn's 2026 Contract Review Benchmark, which tested 11 AI models across 3,282 head-to-head reviews and 21 precision-critical guidelines. The article's broader point is that specialized benchmarks can show how AI behaves inside a structured system, not just in raw model form.
For professional-services firms, that supports a practical approach to custom AI: test the workflow, not just the model. Whether you are building legal document review, accounting intake, or an internal knowledge assistant, evaluation should focus on accuracy, consistency, and how the system performs in the firm's own process.
- Use approved AI and automation to reduce the temptation for shadow tools.
- Build device-change, access recovery, and MFA procedures into any custom workflow.
- Evaluate AI systems as workflows, not just as model names.
- Use benchmarks and internal testing to measure performance on your firm's actual work.
Sources watched
- What Firms Need to Know About MFA When Staff Change Mobile Phones or Devices (CPA Practice Advisor AI)
- AI Adoption Often Slow + Chaotic, TR Survey Finds (Artificial Lawyer)
- The Legal AI Scaffold Changes Everything - Claude Study (Artificial Lawyer)
- What Legal AI Benchmarks Reveal That Model Names Don't (Artificial Lawyer)
