If it seems like everyone is using artificial intelligence (AI) in their work these days, you aren’t imagining it. The recent 2026 AI in Professional Services Report from the Thomson Reuters Institute shows that organizations nearly doubled their use of generative AI in the past year, with more than half of professionals relying on publicly available tools. That means you’re hearing things like “I checked with ChatGPT for ideas” or “Gemini summarized this for me.” Or “Claude wrote all the memos I sent on this contract review project.”
Meanwhile, as world leaders remind people to stay human in the age of AI, professionals are grappling with another balance to strike. They must maintain integrity, protect sensitive data, ensure compliance, and mitigate risk while keeping up with innovation in their businesses and managing time and cost constraints.
While public AI tools may have a place in the routines of many professionals, they increasingly fall short when it comes to delivering legal work. As legal departments handle sensitive and confidential information, the use of inadequately vetted AI can lead to legal errors, reputational harm, and regulatory exposure.
This white paper looks at the dangers of free or general-purpose AI tools and explains the need for legal-specific tools and systems. It also provides a framework for in-house Legal leaders to evaluate legal-specific solutions and best practices for AI adoption in their departments.
The rise of AI for legal work
AI adoption has risen dramatically among professionals since the launch of ChatGPT in 2022. Other general-use large language models (LLMs) followed shortly after, and industry leaders moved quickly to incorporate LLM capabilities into their applications.
The AI in Professional Services report illustrates this dramatic rise. Now, more than 80% of organizations report using generative AI (GenAI) weekly. A full 77% of professionals expect agentic AI to become central to their workflows by 2030. Still, more than a third of corporate legal teams remain unsure about applying agentic AI to their work. There is a very real tension between taking advantage of speed and innovation and making sure the business is fully protected.
General-use tools are at the center of a lot of this tension. They are readily available, cheap or even free, and give answers that are confident, articulate, and filled with specific details. They also make it difficult to verify those responses against their sources or ensure the tools have considered all relevant information.
The good news is that there are AI solutions that significantly reduce your risks and make it easier for you to verify the results. While these tools are not free, they will create meaningful time savings and help you protect your organization while taking advantage of the same types of innovation as your colleagues across the business.
Hallucinations — the risk of fabricated legal details
Hallucinations are one of the largest AI-related concerns. LLMs have a known tendency to agree with the user, meaning they’ll tend to give you the answer you’re looking for, whether the data supports it or not. Consider these recent cases:
- In Mata v. Avianca, a judge sanctioned attorneys for submitting fabricated case law generated by an AI tool. This was the first clear precedent that unverified AI output cannot substitute for legal research.
- In Thomas v. Pangburn a self-represented litigant filed pleadings citing at least ten non-existent cases, burdening court resources.
- Even judges aren’t immune. In the Shahid v. Esaam decision, the judge incorporated AI-generated inaccurate content into an official court order.
An AI platform like the public model ChatGPT is built on a foundation of opaque sources, ranging from public domain books to material scraped from the internet, some of which may be copyright-protected or inaccurate. A lawyer who asks a public AI system for a list of cases relevant to their current task will then have to go over the information they receive with a fine-tooth comb.
A Fiduciary-Grade AI™ tool like CoCounsel Legal from Thomson Reuters, by comparison, composes answers based on its grounding in industry-respected and thoroughly vetted sources, including federal and state legislation and case histories from Westlaw, and know-how guidance from Practical Law.
The need for accurate, trusted legal research and guidance is especially critical as Legal teams are pressured to handle more work in-house. Organizations are now better able to see patterns in the types of matters they face thanks to analytics from Legal Operations practices. They are also getting more precise about which of those matters require outside counsel support and which they want to keep in-house for efficiency and spend management.
If teams take on this work too fast, without the benefit of fiduciary-grade tools, they risk outpacing their own expertise. A free tool or even a startup’s legal-specific tool will absolutely give you an answer about how to handle a new matter. It will state its findings confidently, as all LLMs do. But can you trust that the answer is thorough and accurate? Is it worth risking your company or your department’s reputation to find out?
The legal team at Delta has embraced AI from industry-leader Thomson Reuters, incorporating it into many aspects of its work, including know-how and legal research. “Westlaw and Practical Law have become a go-to resource for up-to-date legal content and precedents. They have deepened our ability to do research on various matters and save significant time preparing for litigation and advising business units as well,” says Laura Dieudonne, Legal Operations and Administration director for Delta.
She notes that using Westlaw and Practical Law has helped the company save more than $200,000 annually by handling more research in-house and insourcing more matters, rather than relying heavily on outside counsel.
“All these tools have transformed the way the legal department operates here,” says Dieudonne. “Thomson Reuters solutions have helped us navigate the complexities of legal operations, keeping us competitive in this fast and ever-evolving landscape.”
Data security and privacy — a critical weakness
Analyzing large sets of data is one of the core uses of AI for in-house attorneys. AI can spot trends and compliance violations in your contracts in seconds rather than the days or weeks it would take a human. It can find inconsistencies between redlined documents and build a checklist for your next merger or acquisition.
But do you really want to hand that sensitive data to a system you don’t trust to keep it private? Are you sure you can trust the answers it gives? Paid LLM subscriptions often tell you that your data is walled off and protected, which you should verify. With public AI tools, you have no guarantees, and they often lack rigorous safeguards for confidential or sensitive information.
That insecurity creates a risk of data breaches, unauthorized access, and regulatory non-compliance. Legal teams must consider this issue through multiple lenses. First, how is your team using AI, and what data are you submitting to LLMs in the course of your legal work? Could a competitor get access to your contract data? Could the market find out about a potential acquisition before you’re ready to announce it?
Second, how are other parts of the business using it? Is the Product team inputting customer usage data to spot trends? What tools is HR using to mine employee data? What Sales data has escaped beyond your safeguards? Most of your colleagues have good intentions, but without properly walled-off and configured tools, they could also be putting your company or customers at risk.
The Legal team at Brinks has struck the balance of innovation and security by using CoCounsel from Thomson Reuters. “If you’re in-house counsel and you have an AI tool like CoCounsel, you have a competitive advantage,” says Hal Snedden, vice president of Global Privacy and Cybersecurity Counsel for Brinks. “Whether you’re negotiating contracts or working with internal functions, AI gives you answers quicker — especially with deep research features.”
The Brinks team doesn’t want to trust their data to the open web. By working with a solution from a trusted source, they can support more deals, improve their work-life balance, and focus on strategic initiatives.
Legal AI solutions — not all are created equal
As AI appears in more aspects of professional life, Legal teams must understand how to select and govern these tools. We’ve discussed the pitfalls of general-use LLMs like ChatGPT. It’s important to note that even tools purportedly “designed for legal” have their pitfalls.
So, what should you look for in AI solutions for your Legal team? Consider the following key evaluation criteria — authority, data privacy and security, human expertise, transparent reasoning, ease of use, and ongoing training and support.
Authority
Look for clarity and defensibility when it comes to data sources. Fiduciary-Grade AI tools must deliver their outputs from authoritative, curated, and domain-specific content, not just information scraped from the open internet. More subtly, they must also operate within the full context required to complete legal work — some tools will confidently cite a dissent as if it were the holding or misrepresent central arguments. Every material output must be traceable to a source that you can locate, cite, verify, and trust. If you can’t easily verify the foundational data for the AI’s output, you risk hallucinations and may be relying on flawed data.
Security certifications and privacy controls
Absolutely non-negotiable is that your legal AI solution must take necessary measures to safeguard your data. Privacy and security must be structural features of the system’s architecture, not policy overlays or configurable options. For example, CoCounsel bases its cybersecurity program on several industry standards, including the National Institute of Standards and Technology (NIST) Cybersecurity Framework and the Capability Maturity Model Integration (CMMI).
You also want to be sure you retain ownership of all your input and output data and that there is a well-defined deletion process.
Human expertise, not just human oversight
As legal providers roll out agentic, multistep workflows, some systems are a black box, from which supposedly correct work product emerges. This output is not sufficient for serious legal work.
Legal workflows must be designed, tested, and continuously refined with meaningful involvement from credentialed lawyers and other subject matter experts. When ambiguity or risk arises, Fiduciary-Grade AI must bring professionals back in rather than generating an output that overstates its reliability.
Transparency
Don’t accept answers from a black box. Be sure to work with a legal AI solution where you have easy access to information about how it works, the process the model uses to answer queries and conduct data searches, and all the steps the provider takes to refine its AI capabilities. AI must be able to provide a reviewable trail of what the system did and what it relied on, sufficient to allow a qualified legal professional — and, where applicable, a regulator, court, or auditor — to evaluate the basis for the output and determine whether the result is reliable and defensible.
Ease of use
Let’s face it, you didn’t go to law school to become a prompt engineer. Look for systems that allow you to pose a single substantive legal question rather than needing to chain together prompts for a result. You want a system that thinks like a lawyer, so you can tell it what you want it to do, and it will work through the project with clear reasoning backed by auditable citations.
Ongoing training and support
AI capabilities and options change quickly. Using AI responsibly and effectively isn’t intuitive. Work with an AI partner that prioritizes regular, intensive user training options for the system. That training may be pre-recorded training modules or access to live support.
Best practices for in-house legal leaders
Leading AI selection and adoption for your department is a big job. You can learn from the experience of others, though. Be sure to plan for assessing and mitigating risk as you build policies for AI adoption and use, to invest in training, and to always keep humans in the loop.
Steps to assess and mitigate AI-related risks
You’ve already shown that you are a thoughtful consumer of AI, simply by reading this far into the white paper.
As you select tools and build governance for AI, consider these steps:
- Connect the use of AI in the Legal department to the AI governance strategy for the larger organization
- Evaluate free and paid tools based on authority, data privacy and security, human expertise, transparent reasoning, and ongoing training and support
- Train your team on proper AI use and share wins to encourage ongoing efficiency gains throughout the team
- Measure efficiencies and track better-than-expected legal outcomes to continue to make the case for Fiduciary-Grade AI tools
Build internal policies for AI adoption and use
Your company’s overall AI policies are a good place to start. Consider going deeper for the Legal team with policies that specify how to use AI. This can mean limiting it in certain circumstances. It may also mean mandating its use, particularly for large data analysis projects that would otherwise bog down an attorney or the whole team.
Train legal teams to recognize and address AI limitations
Be sure your colleagues know the limitations of AI. Here are three tips to have them keep in mind no matter what solution they are using:
- Check citations. Read the cases that your AI solution is recommending. Ensure they are all still good law and that you agree with the way the AI is interpreting them. The legal professional alone is responsible for the cases cited.
- Evaluate tone. Go through every word and make sure it will truly resonate with your reader. AI tools developed for the legal industry produce more contextually appropriate material than consumer-facing options. Even so, you’ll want to calibrate and customize based on intent and audience.
- Ask “What’s missing?” The flip side of hallucinations is missing insight. No AI tool will get everything right all the time, and they can certainly leave things out. Closely edit AI output to be sure your whole point of view is represented and that the language is sharp, crisp, and effective.
Keep the human in the loop
Legal AI tools are not sentient beings. But they might lull you into a false sense that they are authoritative. Good ones offer authoritative sources, sure. They show you where the information comes from. But you are the authority and expert in your business and your legal circumstances. AI can make a great assistant, but it can never be in charge. Keep this in mind throughout your AI journey, and don’t forget that the buck stops with the legal professional, not with the AI tool. Remember the dangers of cognitive surrender, and always keep the final decisions in the hands of human legal professionals.
Free tools don’t offer freedom
In-house legal leaders are walking a tightrope, balancing innovation and security in the adoption of AI across the team. Managing budgets and spend always feels paramount. But the stakes are simply too high to trust your company’s reputation and legal standing to free tools. This is not the place to save money. You’ll always be beholden to the risk that’s inherent in using a free tool that’s drawing from the open web.
Thankfully, there are ways to rigorously assess any technology for accuracy, security, and privacy. With cautious, informed evaluation you can get past assumptions and determine which tools are truly fiduciary-grade. With the right metrics in place, you can demonstrate the value fit-for-purpose tools bring to your Legal team while protecting the organization and your budget.
