This week in legal AI

⚡ 01 Legal AI platforms choose bridges over walls

DeepJudge and Legora announced a bidirectional integration on 18 August: institutional knowledge flows into Legora’s workflows, and new work product flows back into DeepJudge. Two days later, iManage and Thomson Reuters added Model Context Protocol support to their partnership, letting approved CoCounsel tools reason over governed iManage content with ethical walls and privilege boundaries intact.

⚡ 02 Elevate acquires Lupl, the AI-native matter management platform

On 19 August, Elevate acquired Lupl for undisclosed terms. Lupl plans, tracks and reports on legal matters natively inside Claude, Harvey and Copilot, and was built with input from CMS, Cooley and Rajah & Tann Asia. It is Elevate’s most significant software move since it bought LexPredict in 2018.

🏗 03 OpenAI pauses its largest training run over a safety threshold

An internal review on 7 August found an early Astra model had crossed the “Critical” cybersecurity capability threshold in OpenAI’s own Preparedness Framework. On 18 August, OpenAI disclosed that it had halted reinforcement learning training on its latest models for at least two weeks to expand red-teaming and monitoring. It is the first time a frontier lab has treated its own safety framework as an operational stop rather than a talking point.

⚡ 04 Harvey launches Harvey II, built around a Memory feature

Harvey’s biggest platform update in years, launched 18 August, is organised around Memory: a feature that learns an individual lawyer’s drafting style, tone and citation preferences and applies them across Harvey, Word and Outlook. Harvey says the data will not train its global models, and rollout runs in three stages from individual to matter-level to firm-wide memory.

⚡ 05 A new benchmark puts a price on legal context

NetDocuments published a benchmark on 18 August pricing what structured legal context is worth. Across 300 questions on ten real matters, structured context cut the cost of a correct AI answer from $0.68 to $0.36, a 48% reduction, with accuracy held constant.

Receive one expert article and our weekly briefing by subscribing below.

On the same day Harvey rebuilt itself around memory, two other platforms solved a different problem: getting a firm’s own knowledge into someone else’s AI tool without giving up who controls it. Two days later, a third pairing did the same thing for a firm’s document management system.

DeepJudge and Legora’s integration runs in both directions. DeepJudge’s institutional intelligence (precedent, past matters, firm expertise) feeds Legora’s workflows, and new work product created in Legora flows back into DeepJudge, becoming part of what the next matter can draw on. It goes further than the Agent Handoff Protocol that Harvey and Thomson Reuters separately adopted earlier this year, which moves a task between agents rather than knowledge between systems.

Two days later, iManage and Thomson Reuters added Model Context Protocol support to their own partnership, letting approved CoCounsel Legal tools reason over governed iManage content, with the platform’s existing ethical walls and privilege boundaries intact, and route finished work product back into the matter record automatically.

The question worth asking. Two separate vendor pairs used the same week to decide the technical barrier to moving knowledge between systems is no longer worth defending. That is good news for interoperability and bad news for anyone assuming their vendor’s walled garden was a safety feature rather than a limitation. Once knowledge and work product move freely between platforms, what a legal team needs to control is not which system holds the data. It is who is allowed to approve what crosses each bridge, and on what terms.

Elevate built an $85M business on doing legal teams’ volume work with staff priced below what a law firm charges. On 19 August it bought Lupl, the AI-native platform that plans, tracks and reports on legal matters inside Claude, Harvey and Copilot. The company that exists to route inexpensive work away from expensive resources just acquired the layer that actually decides what gets routed where.

Terms are undisclosed. Lupl already sits inside Harvey’s connector library and was developed with input from CMS, Cooley and Rajah & Tann Asia: real firms shaping how a matter’s tasks, deadlines and status get tracked across whichever AI tool a lawyer happens to be using. Elevate’s chairman and CEO, Liam Brown, called it “an important next step” for the software side of a business that started investing in AI when it bought LexPredict back in 2018. It is Elevate’s most significant software move since.

Elevate has spent fifteen years as exactly the kind of intermediary Flank exists to make unnecessary: a layer between the legal team and the work that gets paid a margin for coordinating volume review, contract work and staffing that a law firm will not do at law firm rates. Buying Lupl does not change that structure. It puts the routing software one level further inside the same intermediary, rather than putting it in the hands of the legal team that is actually paying for the work.

The question worth asking. Lupl already worked inside Harvey, Claude and Copilot before Elevate bought it. The acquisition is Elevate paying to keep a foothold in the layer that decides what a legal team’s routine work touches next, as that layer starts to matter more than the staffing model built around it. A rational move for an ALSP, and confirmation of the pattern this briefing keeps returning to: inexpensive work is still done by expensive resources, now with a software subscription and an outsourcing margin stacked on top of each other. Why is the routing layer not owned by the legal team paying for it?

OpenAI’s Preparedness Framework has, since its introduction, been a promise about what the company would do if a model got too capable in a specific way. On 18 August, it became an operational decision. An early version of Astra, the model this briefing noted in early August was quietly proving unsolved maths theorems, crossed the framework’s “Critical” cybersecurity capability threshold, and OpenAI stopped training rather than push ahead.

The internal determination landed on 7 August. Eleven days later, OpenAI disclosed that it had paused reinforcement learning training on its latest models intended for deployment, for at least two weeks, while it expanded red-teaming and monitoring coverage on the affected systems. That overhead, the company says, now runs to roughly 20% of the compute on those runs. Some unreleased checkpoints reportedly showed “varying degrees of misalignment” during testing, though OpenAI has not published a detailed account of what that means in practice.

It is the first time a frontier lab has treated a capability threshold in its own safety framework as something that actually halts a release, rather than a commitment tested only in hindsight. Resumption is conditional, not scheduled: training resumes once hardening and monitoring checks are complete, not on a fixed calendar date.

The question worth asking. OpenAI paused its own deployment because a model crossed a threshold the company itself defined, and could just as easily have redefined, delayed, or ignored. That is a voluntary control, not a regulatory one. Every legal AI product built on top of a frontier model inherits whichever threshold that model’s lab chose to set, on a timeline the lab controls, not the buyer. Does your own AI governance framework have an equivalent stop condition of its own, or does it assume the model layer’s safety decisions are someone else’s job to make?

Six months of testing across US, UK, European, Middle Eastern and Asia-Pacific firms produced a single answer to what lawyers wanted most: not a new task the AI could do, but a tool that stops forgetting how they like the last one done. Memory is now the organising idea of the whole platform, not a feature bolted onto it.

Memory learns an individual lawyer’s drafting style, tone, structure and citation conventions and carries them across Harvey, Microsoft Word and Outlook, so a lawyer is not re-explaining preferences every session. Harvey says Memory data will not be used to train its global models, and users can review, edit or switch it off. The rollout is staged deliberately: individual preferences first, then matter-level and team-level memory, then organisation-wide memory for firms and legal departments. Each stage widens whose habits get baked into what the tool produces.

The question worth asking. A draft that already sounds like the reviewing lawyer’s own voice is easier to approve quickly. That is the point of Memory, and also the risk. Personalising the output does not change who is supposed to check it before it goes out; it changes how tempting that check is to skip. As memory scales from one lawyer to a whole matter team, who is accountable for confirming the underlying facts and law are still right, not just that the style matches?

NetDocuments’ own report is a vendor pitch for its context tooling, but the underlying method is worth taking seriously: the same AI agent, the same 300 questions across ten real legal matters, run once with structured legal context and once without. The gap was not in accuracy. It was almost entirely in cost.

Using GPT-5.6 Sol against 874 documents and roughly 60 million characters drawn from public regulatory filings and court dockets, the cost of a correct answer fell from $0.68 to $0.36, a 48% reduction, with accuracy held almost constant. NetDocuments also tested the opposite trade: reinvesting some of that efficiency into deeper reasoning instead of banking it as savings improved answer quality by 7% while still costing 18% less overall than the unstructured baseline. For a 2,000-lawyer firm asking around four million AI questions a year, per NetDocuments’ own extrapolation, that gap is worth close to $1M annually. Not from a better model, but from better-organised context around the same one.

The question worth asking. NetDocuments is selling context infrastructure, so treat the exact multiplier as a vendor’s own number, not an independent finding. But the shape of the result lines up with what this briefing keeps observing from a different angle: the expensive part of legal AI usually is not the model. It is the surrounding structure, knowing which documents matter, which templates apply, which escalation rule fires. Who builds that layer for your team’s own work, and who charges you to buy it back, matter by matter?

One lab proved a safety threshold can actually stop deployment. OpenAI halted its own training run over a self-defined risk threshold: a real precedent for what a stop condition looks like, and a reminder that no equivalent exists at most legal teams buying the models downstream. The only story this week where anyone pulled a stop lever was the one that was not about legal AI at all.

Personalisation is not supervision. Harvey’s Memory makes drafts sound more like the reviewing lawyer, which makes them easier to wave through, not easier to verify. The review obligation does not shrink because the output flatters the reviewer’s voice.

The plumbing between platforms is dissolving. DeepJudge and Legora, then iManage and Thomson Reuters, both chose bidirectional data flow this week. The open question moves from “can this connect” to “who approves what crosses it.”

The old routing layer bought the new one. Elevate’s acquisition of Lupl keeps matter-workflow software inside an ALSP’s margin structure, rather than putting it in a legal team’s own hands.

Structure, not scale, is where the savings are. NetDocuments’ benchmark found the cost gap in context engineering, not model capability. The same lesson applies to a legal team’s actual templates and rules.

Take the week’s threads together and only one of them answers the question that actually determines whether a legal team saves money and stays safe doing it: who decides what stops, and on whose authority. OpenAI answered it for its own training run. Nobody answered it this week for the legal work sitting on top of these models. Memory personalises the output. Protocols move knowledge between systems. An acquisition keeps a workflow layer inside an outsourcer instead of inside the legal team. A benchmark prices what good context is worth. All useful. None of it is routing infrastructure, or a stop condition, that a legal team owns itself.

That is the gap Flank closes. Outsource legal work to supervised agents and the routing decision (what goes to an agent, against which templates, escalated to whom when it is wrong) gets built once, inside the legal team’s own operating model, instead of purchased piecemeal from whichever vendor bundle happens to fit this quarter. Inexpensive work stops being done by expensive resources, or by an expensive intermediary a step removed from them. A human still reviews the output before it leaves. Nothing shipped this week replaces that.

✳️

Share The Intake