Alternative legal services provider QuisLex today (20 July) unveiled a new advisory framework designed to help legal departments and law firms build the institutional capability needed to produce AI-assisted legal work that can be trusted, governed and defended.
The new offering, centred around an Institutional Capability Assessment (ICA), reflects what QuisLex founder and CEO Sirisha Gummaregula says is the next phase of legal AI maturity. While legal organisations spent much of the past two years experimenting with generative AI tools, the challenge now is proving that AI-generated work can be relied upon and that organisations remain accountable for its outcomes.
“What we’re seeing in the last few months is that using AI has become table stakes now,” Gummaregula told Legal IT Insider. “Everybody is talking about ‘hey, we use AI. We are incorporating it into our workloads.’”
QuisLex has already been active in the AI governance debate through its Five Failure Modes of Legal AI framework, launched earlier this year and submitted to organisations including the ABA, state bars and CLOC. The framework sought to move the discussion beyond hallucinations and highlight a broader set of risks associated with AI-assisted legal work.
However, according to Gummaregula, client conversations quickly revealed a larger challenge. “We started realizing the bigger issue is that you can’t just look at AI using AI in legal departments or law firms by itself. It doesn’t live by itself,” she said. “You really need to build an architecture behind it in the sense that what is your internal knowledge on how you do business, what has already been codified in your templates or in your business practices? Can you make that available to the AI system that is running the work so that it has kind of richer context to that?”
Beyond knowledge capture, Gummaregula says that organisations also need ways to validate outputs and create evidence that supports decisions made using AI-generated work.
“Once the system really executes a work product, do you have processes that allow you to validate those findings and essentially build some kind of data, evidence data on this is the reason why I relied on this work product in case it gets questioned in the future,” she says.
That thinking forms the basis of the new Institutional Capability Assessment. Rather than focusing on a particular technology vendor, the framework is designed to assess and strengthen the client’s own knowledge, processes and governance structures.
“Our new offering is essentially, ‘we will help you build that framework and we are calling it the institutional capability assessment ICA for short,’” Gummaregula said.
“What it allows clients to do is we will use whatever you have already built internally. We’ll codify it in a way that you could use that as the context, as the knowledge context for you to effectively use AI.”
The second element of the framework is reliability and governance. QuisLex says it helps organisations build workflows and controls that allow them to measure and evidence the quality of AI-assisted work.
“On top of the system, we will build methodologies and workflows on how somebody can kind of measure the reliability of what the work that is coming out of the AI system so that you are able to stand behind it,” Gummaregula said.
The significance of the announcement for legal IT leaders is that QuisLex is positioning AI governance as an institutional challenge rather than a technology procurement exercise. While many organisations remain focused on selecting the right tools, Gummaregula argues that technology is only one part of the equation.
“Clients are quite overwhelmed with the options that are out there,” she said. While QuisLex helps clients run proofs of concept and assess technologies, “it is really one piece of the puzzle in a lot of ways… a very important piece of the puzzle. But you really do need to make sure that you’re putting it in the right place, you’re giving it the right context, you’re building the appropriate controls.”
Importantly, the framework is designed to work across different technology environments and industry sectors rather than imposing a single operating model. Gummaregula describes it as “a very organized framework” that helps organisations ask the right questions while remaining highly customised.
“It’s not a one size fits all,” she said. “It’s a framework that allows you to ask kind of the right questions at each level and then it’s customized for that organization.”
Perhaps the most interesting aspect of QuisLex’s thinking lies in how AI changes traditional concepts of liability. According to Gummaregula, organisations are being forced to rethink assumptions that have underpinned legal workflows for decades.
“What does that liability look like?” she questions. “It is a completely different lens now, which is why the industry needs valid framework.” Gummaregula recently wrote an academic paper on this topic and observes: “The workflow is execution, judgment, accountability. We assume that a person who signs the contract is automatically accountable and has exercised judgment. But when you bring in an AI system, execution is by the AI system. Maybe judgment is also – you need to validate whether there was judgment. But the accountability still stays with the individual and not the system. And that changes the entire equation.
“You really do need to look at it holistically on what does the institution needs to do to remain accountable for a work product that they may not have fully generated. Which is why evidencing worse, human judgment appropriately exercised now becomes the bigger part of the equation.”
That shift has led QuisLex to focus heavily on how organisations remain accountable for work products that they may not have fully generated. Human judgment becomes a bigger than ever part of the equation because ultimately, accountability sits with humans. Given the evidence around cognitive surrender – people automatically relying on what their system is saying – it’s important to have plans in place.
“Unless you’re actively thinking about it and making sure you have appropriate controls, you really may be looking at a bigger accountability issue in the future,” Gummaregula says.
