The Pitch Architecture Live preview Glass box What we claim Verify Talk to us
The Pitch

When the regulator asks, you'll have an answer.

AIF wraps cryptographic governance around every AI action your business takes: signed, tamper-evident, on your own servers, with the AI tools you already use. Aligned with the record-keeping the EU AI Act expects from December 2027, and with the tamper-evident logging the EU's Digital Operational Resilience Act (DORA) already requires of financial firms today.

Why this matters now

Three risks already on your board agenda.

Every company in this market faces the same three exposures. Most don't know they're exposed yet.

RISK 01

The audit-trail gap

A regulator, a board member, or a customer disputes an AI-generated decision. You need to produce the exact record of what the AI was asked, what it answered, and which model was used. Chat logs can be edited. Screenshots can be fabricated. There is no cryptographic proof — yet.

RISK 02

The EU AI Act, and DORA already

Article 12 of the EU AI Act requires detailed, auditable record-keeping for high-risk AI in aviation, financial services, healthcare, legal, employment, and critical infrastructure, from December 2027. For financial firms, the EU's Digital Operational Resilience Act (DORA) already requires tamper-evident logging today. Non-compliance with the AI Act carries fines of up to 3% of global annual turnover. Most organisations have no infrastructure for either.

RISK 03

The audit trail lives somewhere else

OpenAI, Anthropic, Google — every AI platform holds the record of what your AI did on their servers, under their terms. If a regulator asks, you're asking a third party for your own evidence. Sovereign jurisdictions and regulated industries can't operate this way.

The clock is running
Dec 2027

The EU AI Act's high-risk Article 12 record-keeping becomes mandatory. For financial firms, DORA's tamper-evident logging already is.

3%
Of global turnover
or
€15M
Whichever is higher
What AIF is

The governance spine your AI runs safely inside.

AIF is not another artificial intelligence. It is the accountability infrastructure that wraps around whatever AI you already use — Claude, ChatGPT, Mistral, Gemini, or a locally-hosted model. Every action your agents take is recorded, cryptographically signed, and linked into a tamper-evident chain. Your existing agents don't change. The audit trail is yours. When the regulator asks, you can answer with proof, not screenshots.

For the engineer
1 URL
Your agents already call an LLM. Change one configuration line. The spine intercepts, governs, and forwards. No code changes. No SDK migration.
For the auditor
0 trust
Anyone can verify the entire audit history using only your published public key. No access to your servers required. No trust in Predict AI required.
For the regulator
Plain English
A regulator asks a question in their own language. The system answers with a cryptographically-signed credential that any standards-compliant verifier can independently check.
What you get

Six outcomes. Each testable by your architect in five minutes.

Every outcome below is backed by running code that 1,341 automated tests verify on every change. Nothing on this page is a "coming soon."

Outcome 01

Every AI action carries a tamper-evident signature.

Built using the same public-standard cryptography governments and financial institutions already rely on. Edit one byte in the audit history and the chain breaks visibly. Any auditor can verify the entire record using only your published public key.

Outcome 02

Your existing agents don't change.

Whether you've built on LangChain, CrewAI, AutoGen, or custom code, one URL change brings every call under governance. No rewrite. No migration. No new SDK to learn.

Outcome 03

Self-hosted.

On your infrastructure. In Europe. Your data, your signing keys, your audit trail. Never SaaS-only.

Outcome 04

Model-agnostic.

Works with Claude, GPT, Mistral, Gemini, and locally-hosted models. Switch providers without re-papering.

Outcome 05

Identity from your IdP.

Azure AD, Okta, Keycloak — we read the identity your corporate login already issues. No new directory.

Outcome 06

Regulators query in natural language.

A regulator asks "show me all decisions involving customer data in Q3." The system answers with the actual events — wrapped as a W3C Verifiable Credential. Independently verifiable by any standards-compliant verifier. Your compliance team's job becomes review, not investigation.

Claims discipline

Equally important — what we don't claim.

Every product claim in this market is a regulatory exposure for the customer. The boundary below is part of why this product is trustworthy.

This is not a compliance certification.

AIF ships the record-keeping mechanics the EU AI Act expects. Certification is your auditor's job — we make their review straightforward.

This does not make AI decisions safe.

We make AI decisions accountable. If a model hallucinates, AIF will sign the hallucination into the chain — and the regulator will see it. The proof is the value.

This is not an AI agent framework.

We do not compete with LangChain, CrewAI, or AutoGen. We govern whatever you've already built on top of them.

The full claims register — every "live" claim, every "in development" claim, every "out of scope" — is published at aif.predictai.ai/claims/.

The math

Priced as a fraction of what it prevents.

Three flat tiers. Unlimited proxy calls. No per-call metering. The enterprise tier costs about 2.5% of your Article 12 fine exposure.

It is the only line item on your board's risk register that shrinks the rest of the register.

STARTER
€1,500/ month
For companies with €10–50M annual turnover
  • Up to 5 governed agents
  • 1 namespace (business unit)
  • Unlimited AI proxy calls
  • EU AI Act Article 12 mapping
  • Email support, 48h response
ENTERPRISE
0.075%/ year
For companies with €200M+ annual turnover · €200K/yr minimum
  • Unlimited governed agents
  • Unlimited namespaces
  • Unlimited AI proxy calls
  • All three regulations + custom templates · multi-instance federation
  • Dedicated engineering line, SLA

Implementation is scoped together — typically an afternoon for the runtime, a week to wire your existing agents in. Managed operations available from €8,500/month for customers who'd rather we run it. See exactly what your tier and exposure look like.

Why now, not later

The deadline moved to 2027. The reason to start now did not.

For financial firms, DORA already requires tamper-evident logging today. And governance tooling takes about a year to procure, integrate, and prove out, so waiting for the December 2027 high-risk deadline means starting the engineering too late.

JAN 2025
DORA in force for financial entities The EU's Digital Operational Resilience Act (DORA) requires tamper-evident logging and third-party risk controls for the ICT systems behind financial decisions. This one is not a future deadline. It applies today.
FEB 2025
EU AI Act prohibitions and AI-literacy duties take effect Article 5 banned-use rules and staff AI-literacy obligations become mandatory.
AUG 2025
General-purpose AI model obligations apply Providers of frontier models attest to risk management and copyright transparency.
AUG 2026
Transparency obligations (Article 50) apply Limited-risk systems must disclose AI use. This part of the timeline held after the 2026 simplification package.
DEC 2027
High-risk record-keeping (Article 12) becomes mandatory Auditable logging for high-risk AI in financial services, healthcare, employment, and critical infrastructure. Moved here from August 2026 by the EU's Digital Omnibus. Procurement and integration lead time is why this work starts now.
AUG 2028
Remaining high-risk obligations fall fully in scope Safety-component (Annex I) high-risk systems reach their compliance date.
Who built this

Predict AI. European-founded.

Built by engineers who have shipped production-grade systems for regulated industries — aviation, financial services, healthcare, and the public sector.

Six of the founder's years were inside Boeing — among the earliest and largest enterprise AI adopters in the United States, where production machine learning has run under aerospace-grade audit, traceability, and certification standards for over a decade. AIF takes that operational discipline — the assumption that every AI decision must be reproducible, signed, and defensible under regulatory scrutiny — and turns it into infrastructure any company can deploy.

We chose self-hosted, EU-built, and audit-first before the EU AI Act was published, because that is the architecture this market always needed. The regulatory tailwind is a coincidence of timing, not a business model.

The product is proprietary. The cryptography is public-standard. The promises are enumerated honestly. Run the four probes against the live demo and verify everything we claim, from your laptop, in five minutes.

Our operating principles

  • Cryptographic first. Every action is signed before it is logged. No exceptions.
  • Intent before action. Purpose is declared and signed before the call goes out.
  • Identity matters. Every agent is a verifiable identity, not a pseudonym.
  • Sovereign by default. Your servers, your vault, your keys, your audit trail.
  • Independently verifiable. The maths is the contract. No trust in Predict AI required.
  • Honest claims discipline. If it isn't on the claims page, we won't tell you it does it.
The next step

Twenty minutes is enough to know if this fits.

A short conversation with your CISO, CFO, or General Counsel — the person who owns AI regulatory risk. We'll scope the deployment, map your agents, and tell you honestly whether AIF fits your situation. If it doesn't, we'll say so.