Legal technology is experiencing a profound transformation, with artificial intelligence (AI) agents taking center stage in compliance, e-discovery, and document review. In the past week alone, several major announcements have underscored just how rapidly the field is evolving, with both established and emerging companies unveiling new AI-powered tools and forging influential partnerships.
On August 27, 2025, Exterro, a pioneer in data risk analysis, revealed a suite of domain-specific AI agents designed to automate and streamline legal compliance tasks. Among the newly launched tools are a Sensitive Data Detection Agent—capable of identifying personally identifiable information (PII)—and a Jurisdiction Mapping Agent that applies complex privacy regulations like GDPR and HIPAA. According to Bobby Balachandran, Exterro’s CEO, "The future of AI is agentic AI and Exterro is leading the way. While the rest of the industry focuses on using third-party models such as OpenAI as a wrapper for their artificial intelligence, or talking about the concept of agentic AI, we are delivering actual solutions that lower risk, reduce cost and increase speed, security, and control." (Artificial Lawyer)
But what exactly does "agentic AI" mean in this context? Exterro was quick to clarify that their system operates with what they call "action-level autonomy," rather than the broader "workflow-level autonomy" that often raises concerns among legal professionals. In essence, Exterro’s agents are empowered to complete specific, well-defined tasks—like data classification, privilege detection, or anomaly flagging—without needing human input at every juncture. However, critical decisions, such as final legal judgments and escalations, always require human oversight. As Exterro explained, "Our agentic system operates with action-level autonomy, not workflow-level autonomy. This is a distinction that needs to be clear and proves crucial in legal and compliance environments." (Artificial Lawyer)
The company’s approach is rooted in transparency and control: "Each Domain-Specific or ‘Expert’ Agent used by Exterro Assist for Data can independently complete specific, well-defined tasks… This comes with an important distinction – they are NOT outsourced to third-party LLMs. These expert agents have legal and regulatory logic embedded directly into their operational models – built by Exterro. However, the overall workflow always includes Human-in-the-Loop decision points." (Artificial Lawyer)
How does this look in practice? When a user specifies an outcome—say, responding to a data breach—the system decomposes the task into subtasks and routes them to the appropriate expert agents. For example, a Sensitive Data Detection Agent identifies PII, a Jurisdiction Mapping Agent applies the relevant regulations, and a Validator Agent reviews ambiguous cases. The system can also autonomously loop, refine, or escalate items if confidence scores fall below set thresholds, but always within "clear, defensible boundaries" and under human supervision.
Importantly, Exterro’s orchestration layer ensures that agents work together seamlessly, handing off tasks automatically and maintaining data within Exterro’s environment. This design, the company says, "ensures that legal and compliance professionals maintain oversight, customer data never leaves the Exterro environment, and Exterro Assist for Data operates in a transparent, controllable, and defensible manner." (Artificial Lawyer)
Meanwhile, another major development in legal AI came from Everlaw, a leading e-discovery company, which announced at ILTACON 2025 that its AI Deep Dive tool was moving from private to open beta. This tool allows legal teams to query massive document collections in seconds using natural language questions. The system then synthesizes answers grounded in facts extracted from specific documents, providing citations and access to the underlying sources for verification. "Deep Dive helps legal teams uncover insights in an entire corpus of data sooner by simply asking questions related to specific issues, parties, or events and get answers in just seconds," Everlaw noted in its announcement.
During a demonstration using a 1.3 million-document dataset from the Mallinckrodt opioid litigation, Everlaw CEO AJ Shankar showed how the tool identifies about 10,000 potentially relevant documents, narrows them down to around 50 key sources, and then extracts and scores facts to provide a synthesized response. The system is designed to be "very transparent to the user," Shankar explained, allowing users to check the relevance scores and review the actual document excerpts supporting each claim. If the evidence within the document set is insufficient to answer a question, Deep Dive will simply state, "No promising answers were discovered." (LegalTech News)
However, Shankar was candid about the system’s limitations: "Does that mean that it gives you great answers every time? Absolutely not… Just as you’d check the facts if an intern or associate brought you something interesting, you want to check the facts here." (LegalTech News) He also emphasized that Deep Dive is not intended for comprehensive document review or privilege determinations, explaining, "I would not use it in relevance or privilege review because those are casting wide nets. You don’t want to miss things." (LegalTech News)
Under the hood, Deep Dive leverages retrieval-augmented generation (RAG) with sophisticated reasoning models, including OpenAI’s o3 model, and partners with Google on advanced embedding models. Documents are processed into overlapping chunks to preserve context, and the system’s focus on narrowing results is intentional: "The bigger the set, the lower the signal to noise ratio of the facts," Shankar said. (LegalTech News)
Potential use cases for Deep Dive include last-minute deposition preparation, early case assessment, and improved planning for document review, according to Chuck Kellner, Everlaw’s senior strategic discovery advisor. The tool has been tested in about 49-50 matters during private beta, with document counts ranging up to 10 million. One user even discovered a previously missed key document during onboarding, highlighting its potential to surface critical insights quickly. (LegalTech News)
Looking ahead, Everlaw plans to enhance Deep Dive with more sophisticated query planning, conversational threading, and expanded context analysis. The company expects to announce general availability later in 2025, though no specific timeline has been set.
In a separate but related milestone, Everlaw also announced that it expects to receive FedRAMP certification in September 2025 for its EverlawAI Assistant suite, covering generative AI features like Review Assistant, Coding Suggestions, and Writing Assistant. This would make Everlaw the first e-discovery vendor to have its full portfolio of generative AI features FedRAMP authorized, enabling federal agencies to adopt its AI-powered platform. (LegalTech News)
The surge in legal AI activity is also reflected in industry collaborations and conferences. On August 27, 2025, Norm Ai, a compliance agent developer, joined Stanford’s CodeX legal tech group to contribute industry perspectives on AI agents in the legal and compliance arena. Roland Vogl, Executive Director of CodeX, welcomed the move, stating, "[Their] participation in CodeX’s affiliate program will provide us with a unique perspective from a leader at the forefront of legal AI as we advance our research in this critical domain." Patrick Vergara, Norm Ai’s COO, echoed the importance of the collaboration: "Norm Ai’s clients face some of the most demanding regulatory obligations worldwide. Working with researchers at CodeX will help us balance innovations in AI agents with the practical realities of legal frameworks." (Artificial Lawyer)
For those eager to stay ahead of the curve, the Legal Innovators Conferences in London (November 4-6, 2025) and New York (November 19-20, 2025) promise to bring together the brightest minds in legal AI for discussions on where the field stands and where it’s heading.
With agentic AI tools, transparent document analysis, and industry collaborations accelerating at breakneck speed, the legal sector is clearly entering a new era—one where human expertise and machine intelligence work hand-in-hand to navigate the growing complexities of compliance and discovery.
