The ProSearch Pulse Volume 12

Legal AI tooling is sprinting ahead of the rules that make its use defensible in court, according to Sidley Austin partner Colleen Kenney. In a Law.com interview tied to Legalweek’s 2026 Leaders in Tech Law Awards, she points to generative AI document review as a clear case where adoption is available in the market, yet still hard to run in discovery.

Key insights

• The main bottleneck for law-firm innovation is the widening gap between rapid tech development and slower legal frameworks, including judicial guidance, best practices, and defensible workflows.

• Generative AI for document review has been commercially available for more than a year, yet firms still lack external scaffolding. Few court orders, negotiated protocols, or model agreements spell out what acceptable use looks like in discovery.

• Sophisticated in-house counsel are pushing outside counsel to use tools that reduce cost, improve accuracy, and increase efficiency. That demand gives firms cover to invest and pilot, especially when they can validate outcomes.

• Firms are moving data and AI leadership into the C-suite, launching firmwide AI initiatives, and building governance frameworks. Training is shifting as well, with new-associate development moving toward practical, AI-native workshops focused on workflows, judgment calls, and defensibility.

Kenney’s forecast combines speed with complexity. Agentic and generative AI could compress timelines for review, privilege calls, early case assessment, and deposition prep. Co-mingled AI- and human-generated data expands preservation and collection questions, and it forces discovery teams to define what “reasonable” means when machines are part of the record.

In the near term, the fastest adoption may land in broadly marketable operational tools like billing and timekeeping, where standards are clearer and buying decisions are simpler.

The long-dismissed idea of “zero backups” is back in the conversation. In Redmondmag, 22-time Microsoft MVP Brien Posey argues that traditional backups now carry growing security and legal liabilities. He also says they remain essential for point-in-time recovery and operational resilience.

Key insights

• “Zero backups” breaks a core requirement: recovery point restores. Replication can also spread Exchange information-store corruption across redundant servers at machine speed unless delayed replicas exist.

• Backup systems have become high-value attack targets. Even without direct access to backup data, a compromised backup console may expose credentials, SaaS tokens, and API access, depending on safeguards.

• If attackers access backup datasets, backups become a data exfiltration and extortion tool. Attackers can resell stolen content or threaten a leak to force payment.

• Retention multiplies legal risk. An organization required to keep three years of data may still hold five years in backups, widening what becomes discoverable in eDiscovery.

The practical takeaway for security and IT leaders is to reduce the backup blast radius. Posey’s framing points toward hybrid approaches: keep backups for recovery, limit retention, isolate admin planes, and harden consoles so backups do not become the easiest path to a breach or expanded legal exposure.

New York lawmakers are advancing legislation that would bar AI chatbots from posing as lawyers and other licensed professionals, and would give misled users a path to sue. The bill’s sponsor, Sen. Kristen Gonzalez, told Reuters that current rules do not clearly prevent large language models from claiming professional credentials and dispensing advice. The proposal moves forward as litigation and regulatory scrutiny around AI tools keeps building.

Key insights

• The bill would prohibit chatbots from providing “substantive responses” that amount to the unauthorized practice of law, using the same standard regulators apply to a natural person.

• The scope extends beyond legal services to other licensed roles, including doctors and mental health providers. Lawmakers point to high-stakes harm when systems generate confident but incorrect guidance.

• Platforms could still face liability even when they disclose users are interacting with a “non-human chatbot,” which weakens disclaimer-only defenses.

• The measure is part of a broader New York AI package, including bills focused on minors’ safety and requirements to conspicuously warn that AI outputs may be inaccurate.

For AI product teams and leaders in regulated industries, the direction is tightening. Consumer-facing chatbots are increasingly treated like accountable service channels. If New York moves first, providers will need stronger identity and role controls, plus tighter professional-content guardrails, to reduce litigation exposure and keep up with state-by-state compliance expectations.

LawFuel reports that the legal AI market is splitting into two lanes, each with a distinct risk profile.

• Authoritative AI, built for citable, high-stakes legal work where grounding, verification, and defensibility matter most
• Operational AI, built for workflow-heavy document tasks where speed and integration tend to win

The timing is telling. Thomson Reuters says CoCounsel reached 1 million users across 107 countries. Anthropic rolled out legal-focused plugins for Claude Cowork.

Key insights ProSearch is tracking

• LawFuel, citing a Fortune commentary by Thomson Reuters CPO David Wong, frames many “AI went wrong” stories as an architecture problem. Tools that lack connections to authoritative legal infrastructure can steer users toward weak or unreliable sources.

• Reliability and liability pressure is rising. Stanford researchers found that even purpose-built legal AI tools generate incorrect information more than 17% of the time. Courts have seen 300+ documented hallucination cases in filings since 2023.

• The market response reflected the competitive shock. LawFuel reports Thomson Reuters shares fell 30%+ after Anthropic’s plugin launch in late January, then jumped 11% after the CoCounsel one-million-user milestone.

• Incumbents are leaning on “content moats” as a durable advantage. Westlaw is described as curated over 175 years, with millions of decisions, statutes, and regulations. It is backed by $200M+ annually in AI investment and about $11B in capital capacity through 2028.

What this means for law firms and legal teams

This reads like a procurement shift. “Legal AI” is no longer a single buying decision.

• Tools aimed at citable research and court-ready outputs will be evaluated on grounding, auditability, and verification workflows. This looks like chain of custody for AI outputs. The reasoning matters as much as the answer.

• Operational copilots will compete on integrations and speed. Professional responsibility still rests with the lawyer who signs the work product.

In the AI + HI model, the technology can accelerate research and drafting, but human judgment remains the control surface. For partners building an AI roadmap, early planning separates a pilot that looks impressive from a system that holds up under real legal scrutiny.