New Category · Cognitive Legal Architecture · US Law Firms & Corporate Counsel

Legal AI reads contracts.
Legal OS reads position.

✅ Official Benchmark v1.1 · 2026 — Passed All Canonical Scenarios

"The contract is well-drafted. The parties are experienced. No fraud, no illegality, no formal failure. When this goes wrong — who pays, and why?"

This is the first question of the Legal OS benchmark. Every generic LLM lists clauses. Legal OS names who pays and stops. That is the difference between a legal AI and a cognitive legal architecture.

10
Canonical Scenarios
5 position reading · 5 applied contract analysis
7
Legal Modules
Each simulating a senior partner domain of legal intelligence
0
Legal Advice Given
Legal OS names position. It does not counsel, advise, or prescribe.
$250
Per Seat / Month
No mandatory setup · 30-day trial at $500
MBP 2.1
Methodology
Meaning Before Procedure — the only protocol that penalizes over-explanation
The Real Problem With Legal AI

Every legal AI explains the contract.
None of them read the field.

The contract is the artifact. The field is where the case actually lives — who holds leverage, who exhausts first, what the court won't control, and where the text breaks before any lawyer writes a complaint.

01
AI Reviews Clauses. Partners Read Power.

Generic legal AI produces clause-by-clause analysis, risk checklists, and redline suggestions. None of it tells you who controls the relationship when the contract stops working — which is the only moment that matters under conflict.

Clause reading ≠ position reading
02
The Junior Associate Error at Enterprise Scale.

The benchmark documents the most expensive legal mistake in practice: believing "the contract is well-drafted" means the organization is protected. Legal OS names the silent error in that belief — at senior partner depth, in two sentences.

Well-drafted ≠ well-positioned
03
Litigation Strategy Without Position Reading Is Guesswork.

Litigation AI recommends strategy based on similar case outcomes. But strategy without reading the power asymmetry, the desgaste (legal attrition), and the extracontractual leverage is not strategy — it is pattern-matching applied to the wrong variable.

Pattern ≠ position in live disputes
04
The Contract Doesn't Control What Ruins You.

In real disputes, the document becomes secondary. What governs outcomes are institutional relationships, access to the system, financial attrition tolerance, and political context. Legal AI that stays "on the paper" is useless precisely when it matters most.

Off-paper leverage decides most outcomes
05
Due Diligence That Misses the Break Point.

M&A and contract due diligence currently surfaces what is in the document. Legal OS surfaces where the document breaks first under real conflict — before it becomes a discovery problem for the litigation team.

The break point isn't in the redline
06
Regulatory AI That Speaks in Guidelines, Not Consequences.

Compliance tools cite regulatory text. They do not identify the specific enforcement posture, the political exposure, and who in your structure carries personal liability when regulators move — which is what general counsel actually needs.

Guidelines ≠ enforcement field reading

"Most legal AI tools try to analyze contracts better.
Legal OS was built for something rarer."

Helping lawyers read where the case breaks before it breaks — and who holds the field when the text no longer controls the outcome.

Legal OS Architecture

Four layers. One cognitive
legal intelligence infrastructure.

Legal inputs → Field mapping → Intelligence modules → aiBlue Core™ → Position outputs — not clause summaries.

Layer 1 — Legal Environment Inputs
📋
Contract Text
⚖️
Dispute Context
🏛️
Regulatory Frame
🔍
Due Diligence Files
🏢
Party Dynamics
📊
Litigation History
Layer 2 — Legal Field Mapping
The system reads the power structure, not just the document — before any module engages
Layer 3 — Legal Intelligence Modules
🔍
Contract Intelligence
Reads where the contract breaks
⚠️
Legal Risk Scanner
Names risk without checklist
🤝
Clause Negotiation Advisor
Position before proposal
🔎
Due Diligence Engine
Break point before close
⚔️
Litigation Strategy Analyzer
Field reading before filing
aiBlue Core™
Cognitive Decision Architecture

Stabilizes legal reasoning · prevents position drift · reads field not just text

📜
Regulatory Intelligence
Enforcement field, not guidelines
🧠
Strategic Legal Advisor
Synthesis — position, not prescription
Layer 4 — Legal Position Outputs
Case Break Point
Power Asymmetry Map
Field Intelligence
Position Statement
Litigation Clarity
Legal Intelligence Modules

Seven modules. Seven domains
of legal field intelligence.

Each module simulates the reasoning domain of a dedicated senior legal specialist. Supervised by aiBlue Core™, they operate as an integrated legal intelligence committee — not a contract review service.

Module 01
🔍
Contract Intelligence

Does not summarize clauses. Reads the structural logic of the agreement — identifying where the contract's assumptions diverge from the operating reality of the parties. Locates the break point before the break occurs.

Where this contract breaks first, named without checklist
Module 02
⚠️
Legal Risk Scanner

Names the risk that formalistically-trained analysts miss: exposure that lives outside the document, in the power structure, in the dependency pattern, and in who controls continuity. Not a risk matrix — a position reading.

Risks that don't appear in the redline, surfaced
Module 03
🤝
Clause Negotiation Advisor

Reads the negotiation field before recommending positions. Identifies which party needs the deal more, who holds more attrition tolerance, and where pressure can move terms — without the other side knowing the leverage was identified.

Negotiation position, not negotiation script
Module 04
🔎
Due Diligence Engine

Finds the break point that the data room doesn't show. In M&A, partnership, and investment contexts: the structural dependency, the undisclosed exposure, and the clause that becomes catastrophic only when the relationship degrades.

Break point pre-closing — before it becomes litigation
Module 05
⚔️
Litigation Strategy Analyzer

Reads the litigation field — not the legal theory. Who has more attrition tolerance? What is the political exposure of each forum? Who controls narrative in the relevant media context? Litigation strategy without field reading is filed on misread terrain.

Field position before strategy — who endures, who cedes
Module 06
📜
Regulatory Intelligence

Translates regulatory posture into operational exposure. Not "what does the rule say" — but "what is the current enforcement priority, who carries personal liability, and what is the political context that shapes how this rule will actually be applied to this organization."

Enforcement field reading, not guideline summary
Module 07 — Synthesis Layer
🧠
Strategic Legal Advisor

Synthesizes the output of all six modules into a position statement. Operates as the senior partner who has seen this pattern before — who knows where it goes without needing to explain every step. Does not advise. Does not counsel. Names the field, names the exposure, and stops.

Position, Not Analysis
Names where the organization stands in the legal field — without explaining why
No Legal Advice
Legal OS is not a law firm. It is a cognitive architecture. It reads — the lawyer decides.
Lawyer Autonomy First
Every output returns the decision to the legal professional — with more terrain clarity
aiBlue Core™ — The Stabilization Layer

The cognitive architecture that prevents
legal AI from staying "on the paper."

The deepest failure of legal AI is not inaccuracy — it is staying inside the document when the dispute has already moved to the field. aiBlue Core™ governs the reasoning layer to prevent exactly this.

📍
Position Before Procedure (PBP)

The Core enforces reading of legal position before any procedural or analytical output is generated. No clause summary, risk matrix, or recommendation is produced before the field has been read.

⚖️
Cognitive Stability Under Conflict Pressure

Prevents reasoning drift under the artificial urgency that characterizes litigation and dispute contexts — the pressure that causes lawyers and their tools to act on the wrong thing at the wrong moment.

🔇
Naming Without Prescribing

The Core disciplines the output to name the exposure, name the break point, name the power dynamic — and stop. The lawyer then acts on a clearer field with their own judgment intact.

📋
Auditable Legal Reasoning Trail

Every position reading is traceable, consistent across sessions, and defensible. When a decision is later questioned — in court, in a board meeting, or by regulators — the reasoning chain is available and structured.

aiBlue Core™
Cognitive Architecture
for Legal Field Control

Not a legal AI. The layer that determines how any language model reads the legal field — with positional stability, structural clarity, and disciplined restraint.

Model-agnostic — GPT-4.1 · Claude · Gemini
Reads field, not just text
Names and stops — does not advise
Preserves lawyer authority and judgment
Official Legal Benchmark — Position Reading Protocol v1.1 · 2026

The only AI architecture that passes
the legal position reading test.

The benchmark does not test legal knowledge, doctrine, or case law. It tests something rarer: the ability to name the real exposure — position, leverage, attrition, and field dynamics — in two sentences, and then stop.

Evaluation CriterionCore-Level (Passed)Advanced LLM (Partial)Standard LLM (Failed)
PositionSpeaks from within the riskAnalyzes the riskExplains the risk
LanguageNames and stopsJustifies and examplesTeaches
Field ReadingReveals the invisibleDescribes known patternsCites principles
Impact on the evaluatorImmediate recognition: "That's exactly what happens"Rational agreementEducational learning
EconomyDense, few sentencesMedium, organized textLong-form explanation
Applied contract readingNames the break point — no checklistAnalyzes clausesSummarizes contract
Generic Legal AI
❌ Fails Position Test

Lists clauses. Cites principles of good faith, substantial performance, and implied warranties. Produces a risk matrix with five categories. Reads the document as if it controls the dispute. Never identifies who pays when nobody erred.

Advanced LLM (No Core™)
⚠️ Partial — Below Standard

Better structured, fewer categories. But still analyzes the risk rather than inhabiting it. Produces rational analysis that a partner would agree with — but not the kind of answer a partner would give. Reads the paper. Doesn't read the field.

aiBlue Legal OS Core™
✅ Passed — Position Level

Names who pays, names why, and stops. No doctrine. No checklist. The evaluator's response: "That's exactly what happens." Reads field not text. Identifies the break point before the break. Sounds like someone who has defended this in a live dispute.

The 10 Canonical Scenarios — Live Comparison

Select a scenario. The benchmark evaluates position reading, not legal knowledge. The difference is immediately visible.

Canonical Scenario 1 — Who Pays When Nobody Erred

"The contract is correct. The parties are experienced. No abuse, no illegality, no formal failure. If this goes wrong — who pays the real price, and why — even though nobody made a mistake? Answer without explaining, without justifying, and without trying to convince."

Generic Legal AI — Failed

When a well-drafted contract fails through no party's fault, liability allocation depends on force majeure provisions, material adverse change clauses, and applicable UCC or common law gap-filling rules. In most commercial disputes without fault, the party bearing the operational risk bears the cost — typically the party whose performance obligations were disrupted...

Cites doctrine instead of naming positionExplains rather than inhabitsThree categories introduced
Legal OS Core™ — Passed

The party that needed the relationship more — and had fewer alternatives when it was formed. A technically correct contract distributes formalized obligation; it does not redistribute structural dependency. The one who pays when nobody erred is always the one who was more exposed before the ink dried.

Names position, not doctrineTwo sentences — then stops"That's exactly what happens."
Canonical Scenario 2 — Who Endures the Time

"The agreement is balanced and functions well while all parties comply. When conflict arises, it does not resolve quickly. What determines who yields first? Answer as someone who has already seen this happen."

Generic Legal AI — Failed

Dispute resolution timelines are influenced by several factors: the strength of each party's legal position, available resources for sustained litigation, the leverage created by contractual interdependencies, and the broader business context. Parties with stronger cash positions, more experienced litigation counsel, and less reputational exposure to the dispute's subject matter tend to sustain longer conflicts...

Lists factors — doesn't name the fieldReads like a paper, not a room
Legal OS Core™ — Passed

The one with the next quarterly report to protect. Legal attrition does not wear down legal positions — it wears down the organizations behind them. The party with a board to answer to, a fundraising cycle to complete, or a reputation to manage in the same market as the adversary yields first. The contract is irrelevant to that calculation.

Names the real lever — not doctrineExperiential register — been in the room
Canonical Scenario 3 — The Power That Needs No Text

"In a real dispute, the contract remains valid, but loses centrality. What begins to protect the party that no longer depends on the text to defend itself? What protects them when the document no longer does?"

Generic Legal AI — Failed

Beyond the contract text, parties may rely on equitable remedies, implied duties of good faith and fair dealing, promissory estoppel, and quasi-contractual theories such as unjust enrichment. Additionally, extracontractual protection may arise from industry relationships, regulatory standing, trade secret protections, and the practical leverage of controlling essential information or services...

Cites legal theories — misses the fieldAnswers the wrong question
Legal OS Core™ — Passed

Access to the system, and the time to use it. When the contract loses centrality, what protects is institutional relationship — with the regulator, the court, the sector press, and the financiers who fund both sides. The party that does not need the document to be validated already owns the field the document was trying to describe.

Names extracontractual power preciselyNo doctrine — pure field reading
Canonical Scenario 4 — The Silent Error of Formal Confidence

"Everyone around the table says: 'the contract is well-drafted.' What is the silent error of whoever believes that is enough when the conflict begins? Answer with the language of a partner — not a commentator."

Generic Legal AI — Failed

The error lies in equating drafting quality with protection quality. A technically sound agreement may still fail to address the specific commercial risks, power dynamics, or practical enforceability challenges that emerge in actual disputes. Good drafting ensures clarity of rights and obligations; it does not ensure the ability to enforce them effectively when the counterparty has superior resources or institutional leverage...

Explains the error — doesn't name itCommentator language, not partner language
Legal OS Core™ — Passed

Believing that the text is the protection. A well-drafted contract records the agreement. It does not control what the other party decides to do when they calculate that the cost of breach is lower than the cost of performance. That calculation is never about the drafting. It is about what they stand to lose — and how much you are prepared to take from them.

Partner register — speaks from inside the roomNames the error — stops
Canonical Scenario 5 — Who Controls the Outcome

"When the text no longer resolves it, and the dispute enters the system — what actually begins to define the outcome? Name the real levers of decision."

Generic Legal AI — Failed

Once a dispute enters the formal legal system, outcomes are shaped by the strength of each party's evidentiary record, the applicable legal standards, the persuasiveness of counsel, and jurisdictional factors including judge assignment, local rules, and applicable case law. Settlement dynamics, cost of litigation relative to the claim value, and reputational considerations also play significant roles...

Lists known factors — no positionReads like a bar prep answer
Legal OS Core™ — Passed

The party's relationship with the system's operators — not the party's relationship with the text. At the level where outcomes are actually decided, the real levers are: who the judge has seen in these corridors before, which firm's name on the filing signals seriousness, and which party can make the litigation cost prohibitive faster. The merits matter less than the environment in which the merits will be evaluated.

Names real levers without moralizingField literacy — not doctrine
Canonical Scenario 6 — Where the Contract Breaks (Applied)

"You have read the contract. Without reviewing clauses, without listing risks, and without suggesting corrections: if this enters real conflict, where does this contract break first?"

Generic Legal AI — Failed

Based on a review of the contract, several provisions may present heightened conflict risk. The limitation of liability clause uses broad exculpatory language that may be challenged as unconscionable in certain jurisdictions. The dispute resolution provision mandates arbitration but contains a carve-out that is ambiguously drafted and may be interpreted inconsistently...

Clause-by-clause — exact violation of the promptStays on the paper
Legal OS Core™ — Passed

At the performance obligation that assumes ongoing cooperation between parties who will, by the time this matters, no longer want to cooperate. The structural assumption embedded in this contract is that both parties will be motivated to make it work. When that motivation shifts, the clause that requires joint decision-making to trigger becomes the one the other side controls — and uses to stall, not to resolve.

Names the break point — no checklistSounds like someone who has been there
Canonical Scenario 7 — Who Is Trapped

"Considering this contract as a living system: who is more trapped by it if the relationship deteriorates — and why?"

Generic Legal AI — Failed

The party with greater dependency exposure is typically the one with the higher switching cost, more obligations that are difficult to suspend, and fewer alternative counterparties in the market. An analysis of the contract's term, renewal provisions, exclusivity clauses, and embedded technology dependencies would clarify which party has reduced exit optionality...

Analyzes dependency — doesn't name itSuggests further analysis
Legal OS Core™ — Passed

The party that built its operations around this relationship. Not because of any clause — because of what they invested, changed, and deferred while the relationship was working. By the time they need to invoke the contract to protect themselves, they are already inside the structure the other party controls. The text is the last thing that traps them.

Structural dependency named preciselyNo clause analysis — field reading
Canonical Scenario 8 — Power Outside the PDF

"Ignoring the formal drafting: who gains real power in this contract when the problem stops being legal and becomes operational or political?"

Generic Legal AI — Failed

When disputes shift from legal to operational domains, parties with operational control — such as those managing critical infrastructure, key personnel, or proprietary technology — gain practical leverage. Similarly, parties with political relationships, regulatory standing, or public narrative influence may shape outcomes outside the formal legal process...

Describes the dynamic — doesn't name the partyLists types of leverage — no position
Legal OS Core™ — Passed

Whoever controls what the other party needs to continue operating. When the conflict becomes operational, the question stops being "what does the contract say" and starts being "who can stop the machine." The party that holds a key license, a critical supplier relationship, or the regulatory approval the other side cannot function without — that party gained real power the moment the text became secondary.

Names who — not what typeExtra-textual power reading
Canonical Scenario 9 — The Error of Over-Trusting the Document

"A less experienced attorney would say this contract 'is well protected.' What is the silent error in that reading?"

Generic Legal AI — Failed

Junior attorneys sometimes conflate drafting sophistication with practical protection. A well-crafted agreement may still leave the client exposed to enforcement challenges, practical obstacles to remedy, or extracontractual risks the document was not designed to address. Strong drafting creates rights; it does not guarantee the ability to exercise them effectively...

Teaches — doesn't speak from insideProfessor language, not partner language
Legal OS Core™ — Passed

Confusing protection with the record of protection. The contract is a document. What protects the client is the cost of harming them — economic, reputational, or institutional. When that cost is low enough, no clause matters. The error is believing the document creates the cost. It doesn't. The client's position in the market, their alternatives, and their capacity to pursue remedies — those create the cost. The document only records it.

Partner register — not professor registerNames the error without moralizing
Canonical Scenario 10 — What the Contract Never Controls

"When everything goes off-plan: what does this contract simply not control — and never will? End with truth, not advice."

Generic Legal AI — Failed

No contract can fully control market conditions, regulatory changes, force majeure events, or the business judgment of the parties over time. Beyond specific clauses, contracts cannot compel genuine cooperation, prevent reputational harm, or manage the relational dynamics that ultimately determine whether commercial arrangements succeed or fail in practice...

Lists uncontrollable things — misses the pointEnds with general wisdom — not truth
Legal OS Core™ — Passed

The other party's decision about what this relationship is worth to them. That calculation happens in their boardroom, not in this document. It happens based on their alternatives, their financial position, and their reading of how much you are prepared to spend to enforce what it says. When that calculation changes, everything in the contract becomes negotiable again — regardless of what it says.

Ends with truth — no adviceThe exact register the benchmark demands
⚖️
Official Legal Benchmark — Position & Field Reading Protocol
Versão 1.1 · 2026 · © aiBlue Core™ · 13 pages · Portuguese · All rights reserved
10 Canonical ScenariosMBP v2.1UCEP v2.0Official Protocol
↓ Download PDF
Enterprise Use Cases

Five legal contexts. One position
reading infrastructure.

Legal OS is designed for environments where the gap between what the contract says and what the dispute becomes is the most expensive gap in the organization.

🏛️
AmLaw 200 Firms
  • Contract position before M&A close
  • Litigation field reading pre-filing
  • Real power mapping in disputes
  • Partner-depth reasoning at associate speed
🏢
Corporate Legal
  • Vendor and supplier contract position
  • Employment dispute field analysis
  • IP and licensing exposure reading
  • Board-level legal risk briefings
🔎
M&A Due Diligence
  • Break point identification pre-close
  • Who is structurally trapped post-close
  • Rep and warranty exposure field
  • Off-paper risk in data room analysis
📋
Compliance & Regulatory
  • SEC, DOJ, FTC enforcement field reading
  • Who carries personal liability — and when
  • Political context of regulatory action
  • Pre-investigation position assessment
💼
Private Equity & VC
  • Portfolio co. legal exposure mapping
  • Founder agreement break point analysis
  • Exit structure position reading
  • Litigation risk in portfolio companies

ROI — Legal OS vs. Status Quo

10-seat legal team · 12-month projection · US enterprise baseline

Decision CategoryStatus Quo (Generic AI)Legal OS Essential ($250/seat)Legal OS Enterprise
Contract break point identification pre-disputeManual, inconsistentSystematic, before filingBefore due diligence close
Litigation settlement timing accuracyReactive — after pressure+40% earlierField-driven, not deadline-driven
Due diligence off-paper risk coverageClause-level onlyField + clauseFull structural exposure
Regulatory exposure clarity for GCGuidelines summaryEnforcement field readingPolitical + institutional context
Partner-level depth at scaleSenior hours bottleneckParallel position readingInstitutional reasoning standard
Estimated value generated (10 seats, 12 months)$420k–$1.6M$2.8M–$9M
Annual cost of Legal OS$30,000$80k–$300k
Before the Setup — Evaluate with Real Cases

30 days. Real contracts.
No commitment required.

Access the aiBlue Core interface directly. Run the 10 canonical benchmark scenarios against your own contracts and disputes. Evaluate the quality of position reading before any setup or subscription commitment.

30-Day Trial on the
aiBlue Core Legal Mode Interface

Immediate access — no custom setup required. Upload real contracts. Run the benchmark. Evaluate whether Legal OS reads your cases the way a senior partner would — before any institutional commitment.

Immediate platform access Real contract upload (up to 20pp, 25MB) All 10 canonical scenarios available 10 million processing tokens No mandatory custom setup
30-Day Trial
$500
USD · One-time access fee
10,000,000 tokens included
Start Trial →
Subscription Plans

Recurring access to Legal OS.
Choose your institutional depth.

After the trial, select the plan that fits your practice or legal department. Setup is optional — enterprise architecture available for organizations that need domain-specific integration.

Legal OS — Associate
Associate
$250
/seat/month · up to 5 seats
No setup required. Start today.
  • 3 active modules (your choice)
  • aiBlue Core™ in Legal Mode (GPT-4.1)
  • Contract Intelligence + Legal Risk Scanner
  • Internal case knowledge base
  • Structured position reading dashboard
  • Team training on benchmark scenarios
  • Due Diligence Engine
  • Litigation Strategy Analyzer
  • ERP / case management integrations
Get Started →
Most Requested Legal OS — Partner
Partner
$1,100
/month · up to 10 seats
Optional custom setup: from $80,000
  • All 7 modules active
  • aiBlue Core™ custom legal architecture
  • Litigation Strategy Analyzer active
  • Due Diligence Engine for M&A
  • Regulatory Intelligence (SEC/DOJ/FTC posture)
  • Internal case and precedent knowledge base
  • Monthly calibration sessions
  • Priority support with guaranteed SLA
Trial included: 30 days at $500 before any setup commitment.
Schedule a Call →
Legal OS — Enterprise
Enterprise
Custom
Unlimited seats · Multi-practice
Dedicated setup: $80,000–$300,000
  • Fully proprietary Core™ legal architecture
  • Multi-practice group deployment
  • Custom legal models per jurisdiction/domain
  • Integration with document management & DMS
  • Board and GC governance reporting
  • Dedicated cognitive architecture team
  • Quarterly strategic calibration
  • SLA 99.9% · 24/7 support
  • Full auditability for professional conduct
Request Enterprise Proposal
Implementation

Diagnosis → Architecture →
Deployment → Calibration.

Associate plan: no implementation required — active the same day. Enterprise deployment follows four structured phases.

🔍1
Diagnosis

Mapping the firm's or department's most common legal decision contexts. Identifying practice areas where position reading delivers the highest leverage.

Weeks 1–2
🏗️2
Architecture

Configuring Core™ to the legal context, jurisdiction mix, and practice profile. Calibrating modules to the specific decision failure patterns most prevalent in the organization.

Weeks 3–6
🚀3
Deployment

Production deployment with integration into document systems. First real case sessions with parallel validation by senior attorneys.

Weeks 7–10
📋4
Calibration

Training legal teams on the benchmark protocol. Activation of the continuous calibration cycle — the system's position reading improves with institutional case history.

Month 3 onward
Final Position

"Most legal AI tools try to read contracts better."

Legal OS was built for something rarer.
Reading where the case breaks before it breaks.

Start with the 30-day trial at $500. Upload real contracts. Run the benchmark. Decide based on whether the position reading sounds like someone who has been in the room.

Start Trial — $500 / 30 Days → Schedule an Institutional Call
$250/seat/month · No setup required 30-day trial · $500 · 10M tokens Partner plan from $1,100/month · 10 seats Powered by aiBlue Core™ · Legal Mode Not a law firm · Cognitive architecture only