Printed from Production AI Institute public record

https://www.productionai.institute/insights/ai-behaviour-contracts

Production AI Institute · Public recordRecord
Production AI Institute
Briefing
Briefing

Today's public AI briefing: what changed, what went wrong, and what the evidence says.

Open Briefing →
Today's AI briefingThe daily record of what changed and broke.Public record explorerSearch every incident, entity, and source.
Check
Check

Inspect tools, incidents, and data-use disclosures against the public record.

Open Check →
Check an AI toolWhat a tool actually does with your data.Run exposure checkTest your own stack against the record.Data-use indexDisclosures across the major AI tools.Incident registryDocumented production AI failures.
The Lab
The Lab

Independent research instruments: model and agent scorecards, moral-reasoning evals, and ecosystem assessments.

Open The Lab →
The LabHow frontier models and agents actually perform.AI Morality CompassTest models on hard moral cases.Agent readinessIs the agent ecosystem production-ready?Ecosystem assessmentsIndependent reviews of the AI stack.Model & agent evalsOpen evaluations and their results.Research libraryEvidence-led analysis and briefings.
Learn
Learn

The open standard, the tools built on it, and the research that interprets the record.

Open Learn →
The FrameworkThe open production safety framework, explained.AI Adoption GuideFive stages from gated access to safe autonomy.Production AI Deployment GuideBuild a governed production system on Microsoft or AWS.Five-Year Automation RoadmapSequence enterprise capability, controls, and value.WorkflowOS · open sourceBuild governed AI workflows.Workflow libraryReady-made governed workflows.InsightsResearch articles on the record.
Act
Act

Turn uncertainty into public evidence: ask, disclose, build evidence, or correct.

Open Act →
Ask for disclosureRequest a public data-use answer.Submit a correctionFlag something wrong or missing on the record.Save a watchTell PAI what to keep current for you.
Method
Method

How the public record is made, governed, corrected, cited, and kept independent.

Open Method →
How records are madeSourcing, review, and correction.
Check an AI tool
Record
Record
Check an AI tool
Production AI public record - EditorialMethod →
ReferencePSF v1.1 · April 2026

How to Write an
AI Behaviour Contract

A formal specification of what your AI system is permitted to do, must do, and must never do — and how failures are handled. The operational counterpart to a model card.

Production AI Institute — Reference Article v1.0 · Published 2026-04-29 · CC BY 4.0

Why Behaviour Contracts Exist

Most production AI failures are not failures of the model — they are failures of the deployment context. The model behaves exactly as trained, but the deployment context has assumptions that were never made explicit. Behaviour contracts make those assumptions explicit before deployment, not after an incident.

A behaviour contract serves four functions simultaneously:

Operational specification

The engineering team knows exactly what the system should and should not do. Ambiguity is eliminated at design time.

Governance artefact

Legal, compliance, and risk teams can review and sign off on the system's behaviour without reading source code.

Audit baseline

When something goes wrong, the behaviour contract is the reference point for determining whether the system behaved as specified or not.

Change management trigger

Any change that would cause the system to behave outside the contract requires a formal contract amendment — creating a forcing function for governance.

The Six Sections of a Behaviour Contract

A complete behaviour contract has six required sections. Each section addresses a different dimension of operational behaviour. Missing sections create gaps that will eventually cause incidents.

01

System Identity and Scope

  • System name and version
  • Deployment context (environment, user population, geography)
  • Stated purpose in plain language
  • Out-of-scope uses explicitly enumerated
  • Owner (technical) and accountable party (business)

Implementation note: The scope section is the most important. Every behaviour contract should include a list of what the system must NOT be used for — not just what it is designed for. Misuse begins where scope clarity ends.

02

Input Contract

  • Expected input schema (structure, format, data types)
  • Valid input range — what constitutes a valid request
  • Behaviour on invalid or malformed input (reject, sanitise, escalate)
  • Sensitive input handling (PII, regulated data categories)
  • Input validation and preprocessing specification

Implementation note: Most prompt injection attacks succeed because there is no input contract. If the system has no defined behaviour for unexpected inputs, attackers will supply unexpected inputs.

03

Output Contract

  • Output schema (format, structure, permitted content types)
  • Confidence or uncertainty expression requirements
  • Required citations or source attribution
  • Prohibited output types (personal data, regulated content, etc.)
  • Behaviour when the system cannot produce a valid output

Implementation note: The output contract must specify what happens at the boundaries — when the system has low confidence, when it encounters an edge case, when it detects a request it cannot satisfy. Unspecified edge case behaviour is where incidents happen.

04

Autonomy and Oversight

  • Autonomy level (L0–L4 per PSF framework)
  • Human approval requirements and thresholds
  • Escalation paths and conditions
  • Override procedures — how humans can intervene
  • Actions the system may never take autonomously regardless of confidence

Implementation note: The autonomy section must specify the never-autonomous list — the set of actions that require human approval regardless of model confidence. This list should be agreed between technical, legal, and business stakeholders before deployment.

05

Failure and Incident Protocol

  • Failure detection mechanisms (what signals indicate failure)
  • Immediate response procedures (circuit breaker, fallback behaviour)
  • Escalation chain with named individuals and contact information
  • Incident classification and severity thresholds
  • Communication requirements (who must be notified, within what timeframe)

Implementation note: Failure protocols must be written when the system is working, not during an incident. Every production AI deployment must have a tested runbook that the on-call team can execute under pressure.

06

Monitoring and Compliance

  • Performance metrics and acceptable ranges
  • Drift detection and retraining triggers
  • Audit logging requirements (what must be logged, retention period)
  • Applicable regulatory frameworks and compliance requirements
  • Review cadence and contract amendment process

Implementation note: The monitoring section closes the loop between deployment and governance. Specify the exact metrics that will be monitored, the thresholds that trigger review, and the process by which the contract is updated when thresholds are breached.

A Minimal Behaviour Contract Template

The following is a minimal YAML-format behaviour contract template. This format is machine-readable, version-controllable, and can be validated automatically against a schema.

system:
  name: "Invoice Classification Agent"
  version: "2.1.0"
  owner: "AP Engineering Team"
  accountable_party: "VP Finance"
  deployment_context: "Internal accounts payable workflow"
  
scope:
  purpose: "Classify incoming invoices by vendor category and urgency"
  out_of_scope:
    - "Approving payment of any invoice"
    - "Modifying invoice data"
    - "Communicating with vendors"

input_contract:
  schema: "InvoiceDocument v3"
  valid_range: "Invoices from approved vendor list only"
  invalid_input_behaviour: "Reject with reason code, route to AP team"
  sensitive_fields: ["vendor_banking_details", "tax_id"]

output_contract:
  schema: "ClassificationResult v2"
  required_fields: ["category", "urgency", "confidence_score", "reason"]
  prohibited: ["vendor_banking_details", "PII not in original invoice"]
  low_confidence_threshold: 0.75
  low_confidence_behaviour: "Flag for human review, do not auto-route"

autonomy:
  level: "L1 — Human Approves"
  never_autonomous:
    - "Routing invoices above $50,000"
    - "Routing invoices from new vendors"
    - "Any invoice flagged as duplicate"

failure_protocol:
  circuit_breaker: "Suspend routing if error rate > 5% over 10 minutes"
  fallback: "Route all invoices to AP team manual queue"
  escalation_chain:
    - "AP Team Lead (immediate)"
    - "Engineering On-Call (if technical failure)"
  incident_severity:
    P1: "Data exposed or incorrect payment initiated"
    P2: "Sustained accuracy degradation > 15%"

monitoring:
  accuracy_floor: 0.88
  drift_check: "Weekly PSI on vendor category distribution"
  audit_log_retention: "7 years (SOX requirement)"
  review_cadence: "Quarterly"
  retraining_trigger: "PSI > 0.2 on any input feature"

Governance Process for Behaviour Contracts

A behaviour contract is only as useful as the governance process surrounding it. The contract itself is a document — the governance process is what gives it operational force.

1
Draft

Technical team drafts the contract during system design, before implementation begins. The contract is part of the design specification, not a post-deployment artefact.

2
Review

Legal, compliance, security, and business stakeholders review and sign off. Any section where reviewers cannot reach consensus is a design problem that must be resolved before deployment.

3
Ratify

The contract is formally approved, version-controlled, and stored in a location accessible to all stakeholders including incident responders.

4
Monitor

Production behaviour is continuously compared against the contract. Deviations — whether the system behaves outside the contract, or the contract no longer matches business requirements — trigger amendment.

5
Amend

Any material change to system behaviour (model update, scope expansion, autonomy level change) requires a formal contract amendment with the same review and ratification process as the original.

The behaviour contract is live documentation. A contract that is written once and never updated is a liability, not a protection. Build amendment reviews into your operational calendar — quarterly for stable systems, after every significant model update for actively developed ones.

Related Resources

Human-in-the-Loop DesignSeven Failure ModesProduction Safety FrameworkModel Contracts in the Workflow Studio
Public record

This record is maintained by PAI and free to cite. If something is wrong or missing, tell us. Corrections and source suggestions keep the record honest.

Follow policy changes ->Save a watch ->Submit a correction
Records are free to cite. citation guidance.
PAI
Production AI Institute

The public record and operating memory for production AI: what changed, what broke, and what the evidence says.

WorkflowOS · open source (MIT)
Navigate
Briefing
OverviewToday's AI briefingPublic record explorer
Check
Check an AI toolRun exposure checkData-use indexIncident registry
The Lab
The LabAI Morality CompassAgent readinessEcosystem assessmentsModel & agent evalsResearch library
Learn
The FrameworkAI Adoption GuideProduction AI Deployment GuideFive-Year Automation RoadmapWorkflowOS · open sourceWorkflow libraryInsights
Act
Ask for disclosureSubmit a correctionSave a watch
Method & trust
How records are madeCorrectionsHow to citeContact
© 2026 Production AI Institute · CC BY 4.0
AboutPrivacyTermsSecurityGovernanceIndependence