AI
Anthropic Thinks Attorneys are Redundant to AI
Alloy Newsletter:
Insights into the intersection of AI, Law, and Business.
Anthropic released their "Labor Market Impacts of AI" report this week, and this chart gave me an uneasy feeling.
It suggests LLMs have the theoretical capability to handle the majority of tasks performed by attorneys. Legal AI is arguably the hottest investment sector across all industries right now. Harvey's $11B valuation at 3.5 years old is Exhibit A. My industry is unquestionably ripe for innovation.
But I don't buy the depth of the theoretical replacement, and a landmark ruling from just last month illustrates exactly why.
In United States v. Heppner, No. 25-cr-00503-JSR (S.D.N.Y. Feb. 10, 2026), Judge Rakoff held that 31 documents a defendant generated using the consumer version of Claude, on his own initiative and without attorney direction, were protected by neither attorney-client privilege nor the work product doctrine.
The practical takeaway for anyone using AI to think through legal strategy on their own: because the consumer version of Claude trains on user data and its privacy policy reserves the right to disclose data to third parties, a user cannot have a reasonable expectation of confidentiality in those communications.
The ruling underscores a structural point: long-standing privilege doctrines govern regardless of the technology used. Attorney-directed use under enforceable confidentiality frameworks may present a materially different analysis.
This is a nuance the "AI will replace lawyers" narrative consistently misses. The value an attorney provides isn't just drafting. It's the privileged channel through which strategy and communication freely flow. AI used under attorney direction, in an enterprise environment with appropriate data rights terms, is a force multiplier. AI used independently by a non-lawyer to develop legal strategy is a liability.
That distinction is why the legal market needs attorneys who understand AI, not AI that replaces attorneys.
