‘Govern the work, not the tool’

In Marlon Hylton’s experience, introducing technology to a regulated profession like law is never just about the shiny new tool. Yet with the generative AI revolution in full swing, much of the conversation hasn't moved past it. There's no shortage of discussion over details — which model, vendor, platform, or feature is optimal — alongside a healthy dose of how these tools may displace the lawyers who use them. Less attention is being paid to more pressing issues like the incentives, the review and validation process, or the professional consequences of error.

These are the questions that matter, and getting to them means adopting a posture of discipline. Legal professionals, Hylton says, must “resist both instincts that tend to dominate this conversation: fear and fascination.” The right questions are always more specific: what task is being done? What information is being used? What duty is engaged? What can be verified and how? And, last but certainly not least, who is prepared to stand behind the result?

“There exists a gap in the discourse and to best fill it, the better unit of analysis is human/lawyer-AI collaboration as a system that gives proper regard to the broader professional context,” says the senior counsel and CEO of INNOV-8 Data Counsel Professional Corporation and INNOV-8 Legal Inc. “Lawyers do not need to reject AI, but they should not pretend that a tool becomes safe because a lawyer briefly reviews the output.”

‘Human-in-the-loop’ is not professional judgment

Hylton has immersed himself in the quandary of AI, researching findings of peer-reviewed studies from other industries and disciplines to write his own paper and teaching a course at Osgoode Hall Law School this fall called Legal Values: Human-AI Systems, Judgment, and Legal Practice. He intends to prepare the next generation of lawyers to chart a thoughtful course when it comes to AI, not simply grab the wheel of a runaway bus.

Hylton frames generative AI as a professional judgment and governance issue. Lawyers have duties to uphold, including competence, confidentiality, and responsibility for work product. Currently, lawyers are assuming that having a person review the output, along with mitigating features like retrieval augmented generation, is enough to meet those bars. It’s not, Hylton warns, because unlike other tools, GenAI’s domain isn’t limited to the background of work.

“The appearance of humanlike cognition and judgment presents real risks of overreliance, confidentiality failures, bias, and false confidence,” he says. “I feel very strongly that ‘human-in-the-loop’ is a big part of ensuring that AI serves law and not the other way around. This is the cross on which I may die in this phase of my career.”

Hylton sees firms focusing on technology adoption and asking secondary questions instead of primary ones, which hinge on professional queries. What tasks are appropriate and which are not? What level of verification is required? What client information may be entered? How do we document the process when the work needs to be defensible?

Firms that persist with the secondary train of thought don’t simply risk the obvious — sensational false citation case headlines, for example — but may end up confusing access with competence. Choosing Harvey and offering some high-level vendor training sessions won’t suffice. Imagine a transcript summary becoming the basis for examination preparation even though no one has checked the passages that were omitted or the underlying documents relied on, or an e-discovery platform that flags privilege and the lawyer treats the label as a legal conclusion rather than a triage signal. The danger isn’t AI’s mistakes, it’s a human’s response to them: that because the tool produces with confidence and polish, a busy professional may lower their guard and defer to it.

Ultimately, when firms get the approach backwards, they fail to recognize and correct for human weakness — precisely what the behavioural science and human-AI research warn about.

“In every technology revolution that we have so far seen, humans have been the weakest link in the chain,” Hylton notes. “We must resist ‘human in the loop’ as a reassurance phrase; GenAI demands the presence of real professional control. Symbolic oversight asks, ‘Did a lawyer review it?’ Meaningful judgment asks, ‘Could this lawyer responsibly verify it, rely on it, and own it?’”

Govern the work, not the tool

Whether AI is here to stay is settled. The work now, especially for lawyers, is laying a governance layer that translates professional responsibility into operating rules. For example, classify risk by task, require independent verification, track court- and client-specific disclosure requirements, and build in escalation protocol. In e-discovery, it should also require familiar disciplines such as validation, audit trails, privilege controls, and defensibility.

“In short, the layer’s job is to make the lawyer’s judgment operational rather than symbolic,” Hylton says, adding that where firms get it wrong is by governing the tool instead of the work.

The distinctions matter because tasks are not interchangeable: Legal research is not the same as transcript summarization; a privilege call is not the same as contract clause extraction; a court filing is not the same as internal brainstorming. Approving a product doesn’t mean professional risk has been solved, and GenAI demands more than generic policies such as “check your work,” which never touch the workflow that produces the reliance in the first place.

“Good governance is a professional control system: task by task, matter by matter, with enough structure that a lawyer can explain what was used, checked, and rejected, and why the final judgment is the lawyer’s own.”

Hylton's caution is grounded in e-discovery, where technology-assisted review earned its place not because a senior lawyer was nominally in the loop but because the profession built disciplined habits around it. Lawyers should borrow that “validation culture” by treating AI outputs as leads, drafts, or triage, he advises, not conclusions.

“That’s the learning I’d carry directly into GenAI because the question is not simply whether the tool is powerful,” he says. “It’s whether the legal team can show why the final work product remains the lawyer's judgment.”

Hylton’s best advice? ‘Let’s cool our jets’

There’s no dispute the new tools are impressive and come with pressure to acquire quickly or be left behind. But the profession is far enough down the road now to see the flip side of the coin: a quagmire of poor adoption, high vendor cost, and frustration from leadership.

So, what's the prudent move?

“Let's cool our jets for a second,” Hylton says. “The most practical first step is to map the work. Then the basic controls become much clearer.”

Applying that governance layer on top — one that is tool-agnostic and guides selection through a task-specific matrix tailored to the firm's own work and clients — is precisely the detailed work most firms have neither the time nor the specialized vantage point to do while also practising law. That’s the gap INNOV-8 was built to fill: moving past the shiny-tool trap by bringing the data, governance, and compliance-risk discipline of its e-discovery roots to how firms manage AI.

As Hylton puts it, “the mature response is neither fear nor enthusiasm. It is process, validation, and professional ownership.”

To learn more about how INNOV-8 can help your firm put that discipline into practice, reach out today.