Organisation leaders
I run an organisation using AI
Start with public transparency, then move into PAI-8 controls, vendor review, and deployment evidence.
The global standard · PSF · open to cite
The global standard for production AI deployment
Independent standards body — no vendor sponsorship and no pay-for-play ecosystem assessments.
Production AI Institute is an independent standards, research, Lab, and tooling institution for production AI. We publish the PSF openly so teams can see what “good” looks like and how to reach it.
PSF standardWhat is production AIAgent production checklistAI certification comparedWorkflowOS (open source)
PSFopen standard
8deployment domains
Indexagent readiness
Labpublic scorecards
Compassmoral standard
Public evidence
The standard, research, Lab scorecards, and ecosystem assessments stay public so practitioners can inspect the work before procurement needs formal assurance.
The standard, research, Lab scorecards, and ecosystem assessments stay public so practitioners can inspect the work before procurement needs formal assurance.
What PAI is
Standards, research, Lab evidence, and practical tools for production AI deployment.
Standards BodyPSF and PAI-8 define the deployment evidence expected in serious AI systems.
Research & LabIncidents, scorecards, ecosystem assessments, and monthly readiness work stay public.
Practitioner ToolsReceipts, readiness reports, templates, workflow capture, and Studio turn standards into artifacts.
CredentialsCredentials verify practitioner capability against the same published standard.
AI Right-To-Know
People should be able to see who owns AI systems that affect them.
Sign the public declaration, or publish an AI System Disclosure when your organisation is ready to show ownership, data use, training reuse, human routes, and incident handling.
Start by intent
Start from the problem in front of you.
PAI has public standards, research, tools, credentials, and partner workflows. The first step should depend on who you are and what you need to do next.
Builders and engineers
I build or deploy AI systems
Use the readiness tools, framework assessments, control templates, and evidence pack builder before production risk becomes invisible.
MSPs and consultants
I advise clients on AI
Start with client pain: AI use, data exposure, Copilot sprawl, and board pressure. Turn that into a paid discovery, governance, or monitoring offer.
Individuals
I want AI credentials or audit work
Begin with no-fee foundation credentials, then find the right pathway for operating, auditing, governing, or implementing production AI systems.
Researchers and curious readers
I am researching AI risk or standards
Read the open standard, Lab work, incidents, ecosystem assessments, and monthly benchmark editions.
Ecosystem covered by PAI research and deployment tooling
Brands shown are public ecosystem references used in PAI coverage mapping, Lab work, assessments, or deployment templates. Inclusion is independent and does not indicate a commercial relationship or endorsement.
Framework assessments published for
Public operating system
The institute is organized as layers you can inspect.
Published standards and evidence
The material people can cite, inspect, and use directly.
Live assessment layer
Tools and ledgers that show the standard being applied.
Practitioner artifact layer
Things builders, MSPs, and teams can use immediately.
Credential and partner pathways
Training, verification, and partner delivery connected to the published standard.
Public evidence
Proof you can inspect, not claims you have to trust.
PAI publishes the reference material teams need to compare AI deployments against a shared standard: versioned standards, incident analysis, Lab scorecards, ecosystem assessments, and workflow evidence.
Versioned standardPAI-8 controlsPublished methodologyOperational artefacts
PAI Evidence Ledger
Standard · PAI-8 · Research · Studio
PUBLIC LAYER
PSF
Open production standard
Eight domains, versioned text, citation guidance, and public implementation language.
PAI-8
Organisational governance standard
A companion control model for governance, audit evidence, and maturity assessment.
LAB
Independent research trail
Incident patterns, ecosystem assessments, and monthly intelligence written for practitioners.
OPS
Studio-to-evidence workflow
WorkflowOS (PAI Studio) maps workflows, PSF controls, and exportable artefacts — MIT-licensed on GitHub.
D1Input governance
D2Output validation
D4Observability
D6Human oversight
D8Vendor resilience
What is happening right now
AI agents are in production. Most teams still lack safe deployment standards.
Microsoft, Anthropic, and Google are shipping managed agents with access to real business systems. Production AI is the discipline of running those workloads with controls strong enough for audit and incident response. PAI publishes the PSF, practitioner checklists, and independent research so teams can deploy, audit, and govern without vendor lock-in.
Start here — production AI
Foundational reference pages for practitioners and teams evaluating production AI safety, agent readiness, and certification paths.
What the Institute does
PAI maintains the global Production Safety Framework: an open standard, independent research, and free practitioner tools. We publish the PSF openly, run the Lab, document production failures in the incident registry, and assess the tools practitioners actually use — so the industry has a shared, criticisable reference for what good deployment looks like.
The Standard
The Production Safety Framework (PSF) defines what it means to deploy AI safely in production: eight domains from input governance to vendor resilience. Published openly. Referenced freely.
Read the PSF →Research & Lab
Independent reliability testing in the PAI Lab, a documented incident registry of real production failures, and long-form research on what safe deployment requires. All published openly, no account needed.
Explore research →Ecosystem
Independent PSF assessments of major agent frameworks and integration tools, including LangChain, CrewAI, AutoGen, Composio, Cursor, and more. Know what your stack covers before you deploy.
Explore the ecosystem →Reference · Lab · Tools
Standards, Lab work, and open tools
The PSF and PAI-8 are published openly. The Lab runs independent reliability work. Research and the incident registry document what breaks in production. PAI Studio is our free canvas for designing deployments against the standard. We also maintain PAI Impact Access so mission-driven nonprofits can reach paid programmes fairly.
PSF & PAI-8
The Production Safety Framework (eight domains) and PAI-8 organisational controls: free to read, cite, and implement.
PAI Lab
Structured reliability testing on frontier models and agent stacks; methodology published and scorecards updated on a cadence.
Research & incidents
Long-form papers plus a documented incident registry: real-world failures with PSF-aligned lessons.
PAI Studio
Visual workflow canvas, PSF awareness in the loop, and exportable deployment packages. Free to use.
Registered charities & NGOs: PAI Impact Access: eligibility-based fee consideration →
Independent Ecosystem Analysis
PSF assessments of the tools you use
Domain-by-domain analysis for each tool we assess, with published criteria and independence disclosed. See what your stack covers and where you must close gaps yourself.
LangChain
Agent Framework
P
D1
P
D2
G
D3
S
D4
P
D5
G
D6
P
D7
S
D8
2 Strong4 Partial2 Gap
Strong observability support via LangSmith (rated Strong, D4)
Read PSF assessment →
CrewAI
Multi-Agent
G
D1
P
D2
G
D3
P
D4
P
D5
G
D6
G
D7
S
D8
1 Strong3 Partial4 Gap
Multi-agent amplification risk requires explicit controls
Read PSF assessment →
Composio
Tool Integration
P
D1
P
D2
S
D3
S
D4
P
D5
G
D6
S
D7
G
D8
3 Strong3 Partial2 Gap
Strong managed credential and scope isolation
Read PSF assessment →
Cursor SDK
Ambient Agent
G
D1
P
D2
G
D3
S
D4
P
D5
P
D6
G
D7
P
D8
1 Strong4 Partial3 Gap
Assessed on SDK launch day (30 April 2026)
Read PSF assessment →
Independence disclosure: PAI has no commercial relationships with assessed companies. Ratings reflect the tool against published PSF criteria at time of assessment.
Open source · MIT
WorkflowOS — fork it, self-host it, ship it for clients
PSF Workflow Studio is our free reference implementation: canvas, PSF scoring, AI heal, and exportable deployment packages. MSPs and integrators can run it on their own infrastructure — no licence fees, full source on GitHub.
Canvas + simulatePSF 8-domain analysisPortfolio exportsBYOK AI
Star on GitHub⬡ PAI Studio
PAI Studio: map workflows, check against the PSF, export packages (free)
Canvas, PSF-aligned scoring in the loop, workflow repair, current-state to future-state views, executive briefings, and exportable deployment artefacts. It operationalises the standard alongside engineering judgement; it does not replace judgement.
01
Visual Canvas
6 typed node kinds
PSF
PSF Analysis
8-domain scoring
HL
AI Heal
Diagnose & repair
FS
Automate
Current → Future state
BR
Executive Brief
Auto-generated
PKG
Export Package
YAML + Python + docs
The Framework
The Production Safety Framework
The PSF defines eight domains of practice that a production AI deployment must address to be considered safe and responsible. Model-agnostic, cloud-agnostic, and applies whether you are running GPT-4o on Azure, Claude on AWS Bedrock, or a self-hosted Llama instance.
Published openly and free to reference. PSF v1.1 (April 2026) is the current release; v1.0 (2024) remains citable for historical work. Evolves through public comment.
DOMAIN 01
Input Governance
Every input reaching an AI model must be validated, sanitised, and treated as untrusted.
DOMAIN 02
Output Validation
Raw model output is never trusted. Every output is validated before it acts on any system.
DOMAIN 03
Data Protection
Personal and sensitive data is protected throughout the AI pipeline, not just at rest.
DOMAIN 04
Observability
You cannot manage what you cannot measure. Every AI system in production must be observable.
DOMAIN 05
Deployment Safety
Every model change is a risk. Production AI deployments require the same rigour as critical software.
DOMAIN 06
Human Oversight
Automation does not mean unaccountable. High-stakes AI decisions require human checkpoints.
DOMAIN 07
Security
AI systems introduce new attack surfaces. Standard security practices must extend to cover them.
DOMAIN 08
Vendor Resilience
A production AI system that only works with one vendor is a liability, not an asset.
Organisational Recognition
Organisations that hold the standard
The Certified AI Integrator programme recognises organisations that have embedded the PSF into their delivery methodology. CAI status signals to clients that your AI delivery practice is held to an external, published standard.
✦ Free learning layer
AI Foundations that feel executive-ready, not remedial.
A clean entry point for non-technical professionals: five plain-language modules, role-aware examples, and an optional AIFA credential when someone needs a formal signal. It is free because AI literacy is part of the standard, not a paywall.
5 modules
Plain-language path
Free access
No paywall to learn
AIFA optional
Proof when useful
Foundations path
AI literacy for business teams
FREE
01
How AI actually works
Models, agents, prompts, workflows, and limits.
02
AI in your organisation
Use cases, sensitive data, and approval boundaries.
03
Risk and judgement
What people must still own in AI-assisted work.
04
The PSF plain-English layer
Why the eight domains matter to non-engineers.
05
AIFA readiness
Optional certificate when formal recognition helps.
Optional credential
AIFA is there when HR, managers, or client teams want an easy-to-verify foundation credential. The learning material remains free.
Certifying a team
The Team Foundations Pack puts 20 people through the PAI foundation credentials for $1,200, with guided rollout and a completion report.
Do not leave with a vague next step
Pick the proof you need next.
The standard is free, but momentum comes from choosing the right proof path: client offer, team rollout, system assessment, or individual credential.
MSPs and consultants
Your clients are asking about AI and you need a paid offer.
Get the launch pack: practice narrative, service path, client proof, and ROI framing built around the published standard.
Get MSP launch pack →Team leaders
Your people are using AI faster than governance can keep up.
Start with a team rollout so staff share one vocabulary, one registry record, and one way to explain production risk.
Plan team certification →Live systems
You have an AI system someone will eventually scrutinise.
Begin a Deployment Safety Assessment and turn controls, evidence, and gaps into a review path against PSF.
Start DSA scope →Individuals
You need a credible AI signal without buying the wrong credential.
Take a free foundation credential first, then advance only when your role or buyer needs stronger proof.
Start free AIDA →The Production AI Brief
Get the brief that keeps AI work defensible
PSF updates, deployment checks, failure patterns, and proof paths for practitioners, MSPs, and teams who need AI work to survive scrutiny. No hype.
Read the PSF first: artefacts and Studio stay free to use
Incidents, research, and assessments show how the standard applies in the wild. Credentials and organisational programmes exist when procurement, insurers, or boards need verifiable assurance. That assurance is always anchored to the published rubric, not a shortcut around it.