At first glance, the announcement feels disruptive.
Anthropic is no longer positioning Claude as a general-purpose AI, but as a tool designed to operate directly within legal workflows.
Contract review.
NDA triage.
Compliance workflows.
Contextual legal briefings.
Standardized responses to recurring legal requests.
These are not marginal use cases.
They sit at the very core of in-house legal operations.
Naturally, one question follows.
If a generalist AI can perform these tasks faster and at a fraction of the cost what remains for LegalTech vendors?
And more uncomfortably: what happens to junior roles traditionally responsible for this first layer of legal work?
The pricing shock… with an asterisk
The entry price point is striking.
Starting at around $20 per month, Claude’s Legal Pack appears dramatically cheaper than most enterprise legal tools.
But in practice, professional use tells a different story.
Once document volume, data separation, privacy controls and enterprise-grade usage are factored in, the effective cost moves closer to $90 per user per month.
Still competitive but no longer “disruptive by default”.
This pricing tension reveals something more fundamental than cost arbitrage.
It forces legal teams to ask a harder question:
What exactly are we paying for?
Where generalist AI still stops
Despite its capabilities, Claude remains a generalist system.
And law is not just about producing text.
Several structural requirements remain largely unmet:
• Auditability and citability of sources
• Traceability of legal reasoning
• Governance and accountability
• Enterprise-grade data security and sovereignty
Anthropic itself is explicit on one point:
all outputs must be validated by qualified legal professionals.
This is not a disclaimer.
It is an admission that responsibility cannot be automated.
What about junior lawyers?
The concern is legitimate.
Tasks traditionally assigned to interns, trainees and junior lawyers document triage, first-level reviews, basic research are precisely those most exposed to automation.
But legal expertise has never been built on execution alone.
As long as AI does not reach 100% legal accuracy and it does not the human role simply shifts:
from production to validation,
from drafting to judgment,
from speed to responsibility.
That responsibility remains non-transferable.
So, is this the end of LegalTech? No, but it may well be the end of LegalTech without substance.
The durable value of specialized legal platforms does not lie in text generation.
It lies in what generalist AI cannot yet guarantee:
• Certified and updated legal sources
• Legal opposability of outputs
• Full audit trails
• Data governance aligned with regulatory constraints
• Security architectures designed for sensitive legal information
These are not optional features.
They are conditions of trust.
Looking ahead
After health, AI is moving decisively into law.
Tomorrow, other regulated domains will follow.
The real question is no longer whether AI can assist legal work it can.
The question is whether we are prepared to delegate responsibility, not just productivity.
And so far, the answer remains clear.
