Bill Gates says legal work could cross AI’s “reliability threshold” within a few years.
I suspect law may cross a more important threshold first.
Software outcomes are often more directly testable than legal judgments. Many consequential legal judgments cannot be reduced to a single objectively verifiable answer. In practice, what matters is often whether a view is accepted and relied on as the basis for what happens next.
Imagine a business team asking AI for a legal view and proceeding on that basis. Later decisions build on that judgment.
Niklas Luhmann, a German sociologist known for his social systems theory, made use of the idea of “uncertainty absorption”: once a decision is made, it becomes a premise for later decisions, while the uncertainty behind the original judgment may no longer be revisited.
At that point, AI is no longer just providing legal information. It is becoming part of how the organization decides.
So the more important question may not be:
“When will AI be as reliable as a lawyer?”
but:
“When will relying on AI-generated legal judgments become normal?”
That threshold may arrive before AI is demonstrably more reliable than human lawyers.
And that is where governance needs to look: not only at whether AI is right, but at where its judgments quietly become premises for what happens next.
https://t.co/Tu2e7E16R6
#LegalAI #FutureOfLaw #AIGovernance
The “junior development” problem may not really be about juniors.
An FT piece on “reinventing the legal apprenticeship” made me think about a slightly different question.
AI is beginning to separate doing legal work from developing judgment through that work.
As AI takes on more of the work traditionally done by junior professionals, it raises an obvious question about employment. But development is a different question.
Why preserve junior work simply because juniors used to learn from it?
Perhaps we shouldn’t.
The better question is what that work was actually developing.
It was never just knowledge.
Research, drafting, review, mistakes and feedback taught professionals to read context: how facts, incentives, relationships, institutional constraints and competing values shape judgment.
In-house, much of the learning came from seeing why a technically sound answer was wrong for this business, this relationship, this moment.
The challenge is not to preserve old tasks. It is to recreate the conditions in which judgment develops.
We used to call that experience.
Now we have to design it.
https://t.co/ATTzeL4ica
I keep coming back to one possibility.
AI could turn legal capability into a source of revenue.
Not because legal judgment suddenly becomes easy, but because parts of that capability may become far easier to scale.
Skadden’s new collaboration with OpenAI points in that direction: AI-powered tools designed to help clients assess regulatory risk and make decisions around transactions and product launches.
The same logic could extend inside a company.
In-house legal teams sit on years of hard-won institutional knowledge: how the business navigates regulation, evaluates trade-offs, takes risk, escalates issues, and decides when to say yes, no, or yes under certain conditions.
Some of that capability could become a paid service, a product feature, or even a joint offering with outside counsel.
Privilege, professional rules, liability and governance would need to be worked through carefully.
But AI may be changing the economics of scaling legal capability.
Legal as a cost center has been the story for decades.
Legal as part of the revenue engine could be a very different chapter.
https://t.co/N4YpRTkRCb
OpenAI and Anthropic are fierce competitors. This week, their CEOs delivered a strikingly similar message to the UN Security Council: frontier AI will require international standards and cooperation.
That is more than a tech story.
What struck me most was the venue.
The Security Council has discussed AI before. Still, there is something different about the leaders of the companies building frontier AI speaking directly to the body responsible for international peace and security.
It suggests that AI is becoming more than a question of innovation, competition or regulation. It is also becoming a question of international security.
That made me think about the nuclear age.
After nuclear weapons appeared, governments gradually built systems around deterrence, non-proliferation, verification and crisis management. AI will clearly require something different. The nuclear order was built mainly around states. Frontier AI is being developed largely by private companies, and the technology is moving much faster than traditional treaty processes.
But the underlying question feels increasingly familiar.
What kind of international framework do we need for a technology whose most serious risks may be beyond the control of any single company or country?
I wonder if the AI debate is starting to shift from how to regulate the technology to where it fits in the international order.
https://t.co/yfsVxqqaWG