Preserving professional standards in advanced AI systems
Highlights
- Human-in-the-loop AI design maps directly to attorneys' ethical obligations under ABA Model Rules.
- Legal AI tools function as supervised assistants, requiring attorney oversight at multiple workflow checkpoints.
- Agentic AI shifts collaboration dynamics while preserving professional control through redesigned human validation points.
As agentic AI capabilities expand across legal practice, a critical question emerges: How much autonomy should we grant these systems? While legal technology grows more sophisticated daily, the answer for legal professionals isn’t found in the capabilities of the AI itself — it’s embedded in the fundamental obligations that define our profession.
Jump to ↓
Beyond best practice: The ethical foundation
The ABA framework: Making oversight mandatory
Transparency as professional safeguard
Preserving professional values in an agentic world
Beyond best practice: The ethical foundation
The legal AI tools attorneys rely on today weren’t designed to operate independently. They were built around human-AI collaboration as a core tenet.
Rather than operating autonomously, these systems are built to be guided by attorneys, highlighting human verification of not just factual outputs, but also underlying reasoning. This human-in-the-loop model deliberately distributes tasks between the legal AI and the attorney, keeping the lawyer meaningfully in control throughout the workflow. That design philosophy resonates strongly with the legal profession, where attorneys broadly believe that AI still requires significant oversight, and tend to think of these tools less as autonomous agents and more as junior colleagues, capable and useful, but not yet trusted to work unsupervised.
This framing is more than intuitive — it maps directly to the work many of today’s legal AI solutions are built for. The majority of legal AI tools on the market focus on tasks that have traditionally belonged to junior attorneys and paralegals: legal research, first-draft generation, due diligence, and discovery review. These are workflows where verification and attorney oversight have always been standard practice, making the transition to AI-assisted work feel like a natural extension of existing process rather than a fundamental disruption.
The ABA framework: Making oversight mandatory
The ethical framework underpinning human oversight goes far beyond preference or risk management. The ABA Model Rules provide the formal foundation for why human oversight is not just best practice in legal AI — it is a professional obligation.
Specifically, Rule 5.1, governing the responsibilities of partners, managers, and supervisory lawyers, and Rule 5.3, addressing responsibilities regarding non-lawyer assistance, together establish that attorneys are ethically bound to supervise any non-lawyer, whether human or AI, assisting in providing legal services, to remain consistent with the Rules of Professional Conduct.
In other words, the human-in-the-loop model isn’t simply a design preference or a hedge against AI error. For lawyers, it’s required.
Human-in-the-loop in practice
The simplest way to understand what human touchpoints look like in practice is walking through how Thomson Reuters’ CoCounsel completes a task. AI does the heavy lifting as an assistant, and humans are given room to exercise their discretion.
Consider CoCounsel’s document review for due diligence use cases. It reads through an entire document set and surfaces instances of a concept like “change of control” through genuine conceptual understanding, pulling relevant information from assignment clauses and related provisions that a keyword search would miss entirely. It’s the kind of task that would take a junior associate days to complete manually.
But CoCounsel doesn’t simply run on its own. Human discretion is built into the process at multiple points, starting before the AI ever begins its work. The attorney selects the specific skill they want to activate — choosing, for example, between a word-by-word review of every document in a set versus a broader trend analysis across a larger dataset. They define which concepts are key to the matter at hand. And because CoCounsel operates as a closed system, the attorney also determines which documents the AI can access — defining the universe it works within.
Transparency as professional safeguard
Once the AI produces its output, the transparency mechanisms become crucial. Every insight CoCounsel surfaces is accompanied by a footnote linking directly to the precise location in the source material it drew from. This isn’t just a convenience feature. It’s a deliberate design choice that makes attorney review fast, targeted, and defensible.
This system functions simultaneously as a verification tool and an audit trail, documenting how each output was generated and giving the attorney a clear line of sight into the AI’s reasoning. Additionally, CoCounsel’s AI Reliability team conducts regular audits of AI outputs against known correct answers, ensuring that model performance remains consistent over time.
The agentic AI challenge
This is, by design, a cycle within which AI plays a highly capable but still supporting role. That’s the hybrid approach most legal professionals are comfortable with today. Agentic AI represents a meaningful shift from that dynamic — moving from AI as assistant to collaborator with a degree of independence that generative tools don’t possess.
And along with that shift, some attorneys are more reticent to use it. That instinct reflects something important about legal culture: control over process is not just a preference — it is a professional value. One attorney captured the sentiment directly, noting that prompting a generative AI assistant and reviewing its output feels right, while an AI that acts without review along the way feels like it crosses a line.
That tension is real, and these are concerns that responsible agentic AI designs must answer. Legal AI companies already rolling out agentic features have taken that concern seriously. They are not removing human oversight but are redesigning what it looks like in an agentic context, embedding human-validated checkpoints throughout autonomous work.
Preserving professional values in an agentic world
The aim is for the core value of collaboration to remain intact — what’s changed is the complexity of the work the legal AI is trusted to carry. As we’ll explore in our next installment, the most successful implementations of agentic AI in legal practice demonstrate that attorney oversight doesn’t disappear — it evolves to match the sophistication of legal technology while preserving the professional obligations that define ethical legal practice.
The question isn’t whether human oversight will remain necessary as AI becomes more capable. The question is how we redesign that oversight to match the power of the tools we’re using while honoring the professional responsibilities that make legal practice trustworthy.
Experience responsible AI collaboration firsthand. See how CoCounsel Legal helps maintain attorney control while amplifying your capabilities. Built specifically for legal professionals with transparency and oversight at every step.
CoCounsel Legal
Work smarter with CoCounsel, seamlessly integrated with Westlaw, Practical Law, Microsoft 365, and DMS partners
Go professional-grade AI ↗
