How many “transformative” AI demos have you seen this year? After a year of helping to drive my firm’s AI integration efforts, I can confirm: the hype often outpaces the reality. But when used thoughtfully, these tools can drive real, measurable change in how we practice law. Here are four critical insights from my experience:

1. LLMs Are Assistants, Not Attorneys

Large language models (LLMs) are trained on vast text corpora where most legal content is encyclopedic (e.g., statutes, summaries, treatises, and secondary sources), rather than fact-intensive analyses. So while they’re great research and drafting tools, the conceptual heavy lifting requires human authorship. The primary arguments and strategies should always be developed by attorneys.

2. The “Perfect Prompt” Does Not Exist

The search for the “perfect prompt” misses a fundamental truth: effective LLM performance requires substantive and often fact-specific guidance for each task. Before an LLM can engage in complex reasoning, it needs a clear briefing and a solid conceptual framework. Investing time upfront to provide analytical context produces far better results than endless cycles of prompt tweaking.

3. Effective Integration Requires Infrastructure Thinking

The greatest AI opportunities in legal practice lie in streamlining administrative work. But realizing that potential requires true systems integration, not just layering new off-the-shelf tools on top of legacy infrastructure. It’s time to move beyond the traditional “vending machine” model of buying a separate app for every task. Instead, empower your own teams to design and automate workflows internally. Tools like N8N, ChatGPT’s Agent Builder, and Claude Skills make this increasingly feasible even for those without coding expertise, provided they’re thoughtfully integrated with legacy infrastructure.

4. Don’t Commit Until You See Commitment

Resist the urge to commit to expensive AI tools before you’ve seen genuine engagement from users. Pilot first: prove that a defined cohort of attorneys or staff actually adopts the tool and integrates it into their workflow. Real adoption, not vendor promises or idealistic hopes, should drive investment decisions.

European Commission Proposes to Ease EU Regulations on Privacy and AI

On November 19, the European Commission is expected to unveil the “Digital Omnibus,” a package of reforms that could reshape the General Data Protection Regulation (GDPR), the AI Act, and the ePrivacy rules. Some of the most significant proposed changes include:

• Excluding certain pseudonymous identifiers from GDPR’s definition of “personal data.” More specifically, if an entity cannot reasonably use particular data to identify an individual, such data is not “personal” in relation to that entity.

• Creating a “legitimate interest” for use of personal data for AI training.

• Giving a one-year grace period to companies that have developed high-risk AI systems.

• Excluding some types of data collection from the “opt-in” requirement: “Currently, websites must get explicit consent before storing or accessing most cookies, think clicking ‘accept’ on cookie banners. Under the proposed changes, companies could collect some data without asking first, either for a limited list of ‘low-risk’ uses or under a broader legal basis called ‘legitimate interest,’ which lets companies argue they can use data if it serves their business.”

For a detailed analysis of the proposed legislation by a privacy watchdog, see this article.

UK High Court’s Ruling in Getty Images v. Stability AI

This ruling addresses one of two major IP infringement lawsuits filed by Getty Images, a global licensor of visual content, against Stability AI, the developer behind the generative image model Stable Diffusion (SD), with the other case proceeding in a US federal court in San Francisco. The British case focused specifically on secondary copyright infringement (the act of importing or dealing with infringing copies), rather than primary infringement (the unauthorized act of scraping content for training), because Getty couldn’t prove SD was trained in the UK.

The High Court ruled that:

• SD doesn’t constitute an “infringing copy” of Getty’s database.

• Stability was liable for trademark infringement because older versions of SD generated images containing Getty’s watermarks, though it has since mitigated that liability by using filtered training datasets and implementing guardrails against prompts requesting watermarked images.

65% of Leading AI Companies Found with Verified Secret Leaks

A comprehensive security analysis of the Forbes AI 50 companies has revealed a systemic cybersecurity crisis, with nearly two-thirds of these companies found to have verified credential leaks in their public code repositories. The exposed secrets include enterprise-level API keys providing direct access to organizational infrastructure, training datasets, and in one case, approximately 1,000 private AI models containing potentially sensitive user data.

According to a recent analysis, there are currently at least twelve pending lawsuits by content owners alleging copyright infringement by AI developers. Therefore, it has been known for some time that AI will present new questions for copyright law. However, a recent lawsuit by Reddit against Anthropic (the company behind the large language model Claude) illustrates that the legal challenges presented by AI extend well beyond copyright issues.

This case notably does not assert copyright infringement. Instead, Reddit’s complaint leverages a set of common law and statutory theories to challenge Anthropic’s alleged unauthorized scraping and use of its user-generated content for training AI models.

There are so many interesting aspects about this case, including the way in which Reddit uses Claude’s own responses to support its contention that Anthropic has not excluded deleted posts from its training data. However, in this post, I want to provide a broader overview of the legal theories behind Reddit’s five causes of action:

1. Breach of User Agreement

Reddit alleges that Anthropic, by accessing and using the Reddit platform, implicitly or explicitly agreed to abide by the terms of Reddit’s User Agreement. This agreement, according to the complaint, prohibits scraping content from the platform and prevents users from commercially exploiting content obtained through Reddit’s services. Reddit alleges that Anthropic breached these specific prohibitions by scraping content to train its AI models for commercial purposes.

2. Unjust Enrichment

Reddit alleges that, by “scrap[ing] and us[ing] Reddit content to train and power a model that has enriched Anthropic to the tune of billions of dollars,” Anthropic “was unjustly enriched at the expense of Reddit.”

3. Trespass to Chattels

Reddit’s complaint includes a claim of trespass to chattels, which traditionally applies to interference with tangible personal property. In this digital context, Reddit alleges that its platform, including its servers and technological infrastructure, constitutes its personal property. The complaint contends that, by allegedly engaging in unauthorized scraping, Anthropic intentionally interfered with Reddit’s property, thereby diminishing its server capacity and functionality for legitimate users.

4. Tortious Interference with Contract

The fourth legal theory focuses on Anthropic’s alleged disruption of Reddit’s existing contractual relationships with its user base. Reddit maintains that it has established contractual commitments to users regarding privacy protections and content control mechanisms. Reddit alleges that Anthropic deliberately interfered with Reddit’s ability to honor its commitments to its users by scraping data without establishing proper licensing arrangements that would include sufficient safeguards.

5. Unfair Competition under California Law

Reddit’s final claim leverages California Business & Professions Code Section 17200, a comprehensive statute designed to combat unlawful, unfair, or fraudulent business practices. Specifically, Reddit alleges that Anthropic’s conduct constitutes unfair competition in the marketplace. Reddit argues that these actions resulted in the misappropriation of valuable content, posed threats to Reddit’s platform value, and created competitive disadvantages in the market.

Despite these procedural hurdles, the nearly unanimous Republican support for the AI moratorium initially appeared to signal a remarkable party consensus on a major issue of technology regulation—one with far-reaching implications for federal-state balance in the tech policy arena.

However, recent public statements from GOP senators and Republican attorneys general suggest that this perceived Republican unanimity may have been overstated.

GOP Senators Voice Concerns

The cracks in Republican support became evident during recent Senate hearings and public statements. At a May 21 hearing on AI impersonations, Senator Marsha Blackburn noted that her state of Tennessee “certainly needs protections” from AI impersonations, adding that “until we pass something that is federally preemptive, we can’t call for a moratorium.” It is interesting that Blackburn’s statement defends state-level some AI regulation on their merits, and not just because of federalism concerns.

Blackburn’s statements followed earlier statements by Senator Josh Hawley of Missouri. In an interview, Hawley expressed concerns about the moratorium both because of federalism concerns and because of his belief in the necessity of some AI regulations: “I would think that, just as a matter of federalism, we’d want states to be able to try out different regimes that they think will work for their state. And I think in general, on AI, I do think we need some sensible oversight that will protect people’s liberties.”

Bipartisan State Opposition

Perhaps the most significant challenge to the moratorium came on May 16, when 40 state attorneys general—including 14 Republicans—sent a letter to Congressional leaders opposing the moratorium.

The attorneys general argued that such sweeping federal preemption would undermine states’ traditional authority to protect residents from emerging AI-related harms, including deepfakes, algorithmic discrimination, and data privacy violations. They warned that the moratorium would leave consumers vulnerable while Congress remains slow to enact comprehensive federal AI legislation:

Imposing a broad moratorium on all state action while Congress fails to act in this area is irresponsible and deprives consumers of reasonable protections. State [attorneys general] have stepped in to protect their citizens from a myriad of privacy and social media harms after witnessing, over a period of years, the fallout caused by tech companies’ implementation of new technology coupled with a woefully inadequate federal response. In the face of Congressional inaction on the emergence of real-world harms raised by the use of AI, states are likely to be the forum for addressing such issues. This bill would directly harm consumers, deprive them of rights currently held in many states, and prevent State [attorneys general] from fulfilling their mandate to protect consumers.

The letter urged Congress to reject the proposed language and instead pursue collaborative federal-state approaches to AI governance:

To the extent Congress is truly willing and able to wrestle with the opportunities and challenges raised by the emergence of AI, we stand ready to work with you and welcome federal partnership along the lines recommended earlier. And we acknowledge the uniquely federal and critical national security issues at play and wholeheartedly agree that our nation must be the AI superpower. This moratorium is the opposite approach, however, neither respectful to states nor responsible public policy. As such, we respectfully request that Congress reject the AI moratorium language added to the budget reconciliation bill.

Industry Opposition

This isn’t an example of opposition to the moratorium by Republican officials, but I wanted to include a reference to the recent op-ed by the Anthropic CEO Dario Amodei. Anthropic, as many of you know, is the company behind Claude — a Large Language Model with reporedly 18.9 monthly active users worldwide. In his piece, Amodei opposed the moratorium and instead called for federal regulation of the AI industry.

While acknowledging that the “motivations behind the moratorium”—namely, “prevent[ing] a burdensome patchwork of state laws [that] could compromise America’s competitive position against China”—are “understandable,” Amodei argued that “a 10-year moratorium is far too blunt an instrument”:

A.I. is advancing too head-spinningly fast. I believe that these systems could change the world, fundamentally, within two years; in 10 years, all bets are off. Without a clear plan for a federal response, a moratorium would give us the worst of both worlds — no ability for states to act, and no national policy as a backstop.

“Instead of a moratorium,” Amodei called for federal “transparency standard for AI companies, so that emerging risks are made clear to the American people”:

This national standard would require frontier A.I. developers — those working on the world’s most powerful models — to adopt policies for testing and evaluating their models. Developers of powerful A.I. models would be required to publicly disclose on their company websites not only what is in those policies, but also how they plan to test for and mitigate national security and other catastrophic risks. They would also have to be upfront about the steps they took, in light of test results, to make sure their models were safe before releasing them to the public.