Cut through the hype. Here's what in-house legal teams are really doing with AI and where the limits still are.

‍

Ready to go beyond the basics? Join our free masterclass on June 4th: AI in Legal: What Actually Works (and What Doesn't) and see how in-house legal teams are building real AI workflows today.

‍

Artificial intelligence has moved from the sidelines into the daily workflow of in-house legal teams. Almost every general counsel is now using some form of AI, and most legal departments are no longer asking whether to adopt AI. They're asking how to do it well.

But the gap between what AI promises and what it actually delivers in legal practice is still wide enough to cause real problems. Hallucinated citations have ended up in court filings. Confidential data has been exposed through consumer tools. And a large share of legal teams that say they use AI are still just using ChatGPT in an unstructured, ungoverned way which is a far away from a proper AI workflow.

This post breaks down what's actually working, what the tool landscape looks like, what's genuinely not possible yet, and what's blocking adoption for most teams.

Why Legal AI Adoption Is at an Inflection Point

The pace of change in the last twelve months has been striking. AI adoption among in-house legal professionals has more than doubled in a single year: from a minority of teams experimenting, to the clear majority actively using it. Europe is actually leading this charge globally, with in-house legal teams on the continent showing higher adoption rates than their counterparts in the US.

But here's the nuance that most headlines miss: adopting AI and deploying it effectively are two completely different things. The majority of legal professionals are still relying on general-purpose tools like ChatGPT or Microsoft Copilot for everyday tasks which is fine for writing emails and summarizing documents, but falls short for anything that requires legal precision, privilege protection, or jurisdictional accuracy.

At the same time, a growing number of purpose-built legal AI platforms are maturing rapidly and for teams willing to invest in the right infrastructure, the efficiency gains are real and measurable.

The Tool Landscape: From General to Specialized

Understanding the AI tool landscape for legal counsel means thinking in layers, from the most general to the most specific.

General-Purpose LLMs

Tools like ChatGPT, Claude, Google Gemini, and Microsoft Copilot sit at the broadest layer. Almost all legal professionals have touched these tools. They're useful for drafting correspondence, summarizing documents, structuring arguments, and getting a first draft onto the page.

The critical caveat for European in-house counsel: using a consumer-grade version of any of these tools to process confidential client information or privileged communications is a legal risk, not just a policy question. Submitting information to a public AI platform can constitute a third-party disclosure — destroying attorney-client privilege. Enterprise versions with proper data processing agreements and zero data retention are a different matter, but the distinction needs to be understood before any tool is deployed.

Legal-Specific AI Platforms (Horizontal)

This layer is where purpose-built legal AI lives. These are tools trained on legal content, grounded in verified databases, and built with the professional obligations of lawyers in mind.

Legora (Sweden) has become Europe's best-funded and most prominent legal AI platform, built for research, document review, and drafting across multiple jurisdictions. It's particularly strong for large European law firms and legal teams operating across borders. Noxtua (Germany) takes a different approach: partnering with century-old European legal publishers like C.H. Beck and operating on sovereign German infrastructure, making it particularly relevant for teams where data residency is a hard requirement.

On the global side, Harvey AI dominates in large US law firms, while Thomson Reuters CoCounsel and LexisNexis Lexis+ AI bring the advantage of being grounded in massive, verified legal databases which are significantly reducing (though not eliminating) the hallucination risk that plagues general-purpose tools.

Contract Lifecycle Management (CLM) with AI

For in-house teams, contract management is often the highest-volume daily workflow — and it's where AI delivers the most immediate, measurable value.

Luminance (UK/Cambridge) is the European contract AI platform with the longest track record, originally built for M&A due diligence and now covering the full contract lifecycle. It recently integrated with LexisNexis to bring citation-backed research directly into contract review workflows. Juro (UK) is widely used by European scale-ups and mid-market teams for its browser-native, easy-to-use approach. Tomorro (France) stands out for multilingual contract management, particularly relevant for French-speaking teams and cross-border European work. Legartis (Switzerland) is purpose-built for in-house counsel reviewing incoming third-party contracts. They are GDPR-compliant, ISO 27001-certified, and hosting all data in Switzerland and Europe.

GAIA (Germany) takes a different angle that's worth understanding: it's positioned as a full legal operating system rather than a pure CLM. It covers the entire contract journey: Creating, reviewing, signing, storing, and analyzing agreements, but its standout feature is enabling non-legal teams like finance and HR to handle routine legal tasks independently, within guardrails set by the legal department. For in-house counsel who spend a disproportionate amount of time fielding low-complexity requests from the business, GAIA's cross-functional model addresses a real pain point. It is GDPR-compliant and ISO 27001-certified, making it a credible option for European legal departments where data residency and security standards are non-negotiable.

US-origin platforms like Ironclad have a strong European presence too, and are worth evaluating for larger enterprise deployments.

‍

Interested to see GAIA in action? 

‍

Specialized: Contract Review and Due Diligence AI

Narrower than a full CLM, these tools focus specifically on reading and analyzing contracts at volume. Kira Systems (now part of Litera) is used by the majority of the world's top M&A practices for due diligence. Robin AI (UK) and Definely (UK) serve in-house teams with contract review and drafting assistance. Luminance also operates here. GAIA is worth noting in this category too: its agentic contract extraction capability automatically pulls key terms, dates, obligations, and risks into structured data and its email agent feature allows contracts to be automatically received, analyzed, and returned with a comprehensive review without manual uploads. For in-house teams dealing with high contract volumes, that level of automation in the intake and extraction workflow is a meaningful operational advantage.

The efficiency gains in this category are among the most well-documented in all of legal AI. Contract review time can be reduced dramatically, and the tools are good enough that most teams can justify the investment within months.

Legal Research AI

For research grounded in actual case law and statutes, the key tools are Thomson Reuters CoCounsel (Westlaw-grounded), LexisNexis Lexis+ AI (with strong European coverage), and vLex / Vincent AI (Spain-origin, strong for multi-jurisdictional and cross-border European research). Noxtua covers German-language and DACH-region legal research specifically. Silex (Switzerland) offers a privacy-first research platform for Swiss and European legal professionals.

eDiscovery and Legal Operations

Relativity remains the global standard for eDiscovery, with significant cost reductions now achievable through AI-assisted review. For legal operations and outside counsel spend management, Brightflag (Ireland) is the standout European-origin platform, offering AI-powered invoice review and spend management with a strong EMEA client base.

What Actually Works: The High-ROI Use Cases

Across all the research and practitioner evidence, a clear set of workflows emerges where AI consistently delivers real value for in-house teams.

Contract review and NDA triage is the single highest-ROI use case. The pattern recognition and deviation analysis that AI does well maps almost perfectly onto the repetitive, high-volume work of reviewing incoming third-party paper. Teams that have automated NDA triage — flagging the known deviations and routing only the genuinely novel issues to a human reviewer — report freeing up significant attorney time every week.

Legal research works well when the AI is grounded in a curated legal database and the question is within a well-documented area of law. It's a powerful first-pass tool that surfaces relevant precedent, generates research memos, and identifies potential arguments, as long as a lawyer verifies everything before it goes anywhere.

Document drafting and summarization are the most accessible entry points. AI doesn't write a final brief or a complex commercial agreement on its own, but it dramatically reduces the time to get from a blank page to a solid working draft.

Due diligence at scale: reviewing large volumes of documents in M&A or regulatory investigation contexts is genuinely transformed by AI. Work that previously required teams of lawyers reviewing documents for weeks can now be done in days.

What Doesn't Work (Yet): The Real Limits

Understanding what AI cannot do is just as important as understanding what it can  and in legal work, getting this wrong has professional consequences.

Hallucination is structural, not a bug. AI generates plausible text, not verified legal truth. Even the best legal AI platforms hallucinate sometimes. The risk is highest in less-documented areas of law, local jurisdictions, and emerging regulatory questions. Hallucination rates in local jurisdiction research can be alarmingly high and in some tested scenarios, completely unreliable. Every AI output that contains legal citations or factual claims must be verified before it is used.

Novel law and emerging regulation are outside what any AI can reliably handle. The EU AI Act, new ESG disclosure frameworks, evolving data protection requirements: these are areas where AI can help structure your thinking but cannot provide reliable answers. AI looks backward; it cannot reason about questions where the law is still being written.

Cross-document and complex multi-party analysis remains genuinely hard. AI has technical limits on how much it can hold in mind at once. Analyzing obligations and risks across a complex deal structure with dozens of interdependent agreements, or running privilege analysis across a large document set, still requires significant human oversight.

Jurisdiction-specific work below the level of major courts is a persistent weakness. AI tools are trained heavily on high-profile, well-documented law. The further you go into local regulations, lower court decisions, and niche practice areas, the less reliable the output becomes.

The first-draft trap is a subtler but real risk. Once a lawyer sees an AI-generated draft, it anchors their thinking. Options the AI didn't generate don't get evaluated. In legal work, seeing what isn't there is often the whole point. For instance, the risk that isn't flagged or the clause that's conspicuously absent, AI cannot see what it doesn't generate.

Negotiation, judgment, and strategic counsel remain fundamentally human. AI can prepare you for a negotiation. You can use it for benchmarking terms, flagging deviations or modeling scenarios. But the table dynamics, the psychological choreography, the judgment about when to hold and when to concede, and the relationship context that determines what a client actually needs and none of this is automatable. These are the areas where experienced in-house counsel's value is highest and least replicable.

Accountability is non-delegable. AI cannot be sued for malpractice. It cannot be sanctioned by a bar association. It cannot be held accountable for the advice it generates. Courts have been unambiguous: the professional responsibility for any AI output rests entirely with the lawyer who uses it. This isn't changing.

What's Blocking Adoption and How to Move Past It

For the large share of legal teams that are stuck between experimentation and genuine deployment, the blockers tend to cluster around a few recurring themes.

Trust in output quality is the most commonly cited concern. It's legitimate and the answer isn't blind faith in any tool, but building verification into the workflow. The goal isn't AI that's never wrong; it's a process that catches errors before they cause problems.

Data privacy and confidentiality are real concerns, especially for European teams operating under GDPR. The answer is choosing tools with proper data residency, zero-retention agreements with underlying model providers, and clear contractual protections. Simply avoiding AI altogether is not an option anymore.

The governance gap is perhaps the most consequential blocker. A significant majority of legal teams that use AI have no formal policy governing how it should be used. Without policy, lawyers default to whatever tool is easiest which means often a consumer-grade LLM on a personal account. This creates the very risks that governance is supposed to prevent. Getting a basic AI use policy in place is foundational.

Senior buy-in and cultural inertia remain real in many organizations. The pattern that shows up consistently: the most enthusiastic AI adopters are mid-career lawyers and legal ops professionals; the resistance tends to come from senior partners and leadership. The most effective approach is to start with a specific, high-volume pain point. For instance, NDA triage is the classic first use case. It is able to demonstrate measurable time savings, and lets results build momentum.

The Bottom Line for In-House Legal Teams

AI is not going to replace lawyers. Most likely, it will eliminate mediocrity supported by inefficiency and that's a meaningful distinction.

The in-house legal teams that are pulling ahead are not the ones that have deployed the most tools or spent the most on technology. They're the ones that have made deliberate choices: identifying the two or three workflows where AI genuinely reduces burden, choosing tools appropriate for those workflows, building the governance to use them responsibly, and developing the internal capability to evaluate AI output critically.

The gap between individual experimentation and institutional deployment is the central challenge of legal AI right now. Closing that gap and moving from "some lawyers using ChatGPT on their own" to "a legal department with a coherent AI strategy" is where the real work, and the real competitive advantage, lies.

That's exactly the conversation this masterclass is designed to have.

‍

Join us on June 4th, 2026 for our masterclass: AI in Legal — What Actually Works (and What Doesn't). Register now to secure your spot.