IRIA, the AI Resistance Index

Could I, alone and with an AI, rebuild your product without ever seeing its source code?

Since June 2025, I have put this question into every technical audit I run, in the form of an index I created. If the answer is yes, the intellectual property in the code no longer protects anything, and the company is exposed to new entrants.

IRIA scores what still protects the company, namely five moats an AI cannot regenerate. Each one gets a score from 0 to 4, and the overall index reads out of 20.

Book a scoping call
IRIA radar: the five moats scored from 0 to 4Five-axis pentagon (certifications, partnerships, service guarantees, data and sovereignty) graded from 0 at the centre to 4 at the edge. The plotted profile is fictitious, with 1 for certifications, 2 for partnerships, 1 for service guarantees, 3 for data and 2 for sovereignty, 9 out of 20 in total.1234CertificationsPartnershipsService guaranteesDataSovereignty
Fictitious profile, for illustration, at 9 out of 20. The shortest axes, here certifications and service guarantees, point to the moats a new entrant would cross first.

Why code no longer protects you

The cost of producing software is collapsing. I built Qualneo, a management tool for Qualiopi-certified training organisations, in two months and part-time, between two other projects, for a few hundred euros of AI subscriptions.

I rebuilt it from scratch, without access to anyone's code. The features of a piece of software cannot be protected by copyright (CJEU, SAS Institute, 2012), while its code can. If I can do it between two projects, any team equipped with AI can do it too.

Implementation, meaning the code, is gradually becoming secondary. What makes a company valuable, and what protects it, lies in my view in five moats. None of them sits in the code repository, and that is why they hold.

The five moats

I score each moat from 0 to 4 on evidence, based on what the code, the servers and the interviews show. For each one, the report details the observed state and how to strengthen it.

01

Certifications and standards

What it protects

A certification or qualification (ISO 27001, HDS for health data hosting in France, SecNumCloud) is earned over months of audits, written procedures and evidence validated by a third party (an accredited certification body or ANSSI, the French cybersecurity agency). GDPR compliance demands the same long-term work of proof. An AI generates code in seconds, but it produces neither the audit file nor the signature of the certifying body. A new entrant can copy the feature, but it cannot copy the auditor's stamp.

What I look at to score it

I look for a certification issued by a third party, and for GDPR compliance that shows in the product itself. Compliance that only exists in documents pulls the score down, even when the data is sensitive. A regular penetration test is a good start, though it does not carry the weight of a third-party stamp.

How to strengthen it

  • Bring GDPR into the product with a data protection impact assessment (DPIA) where the processing requires one, a legal basis, encryption at rest and a controlled retention period.
  • Use it as the first step towards a certification that fits the data you handle, such as ISO 27001, or HDS where the product hosts health data on behalf of others.
02

Locked-in partnerships

What it protects

A locked-in partnership is access or distribution negotiated and written into a contract with a third party, be it a key customer, a hardware manufacturer or a reseller network. No AI negotiates an exclusivity on your behalf, and a new entrant has to go through the whole commercial and human journey again to get one. This moat is dug on commercial ground, where code has no grip.

What I look at to score it

I separate the real commercial relationship (reseller network, white label, hardware installed at customer sites) from the lock written into a contract. A relationship that makes switching costly for the customer already counts. A signed exclusivity counts more.

How to strengthen it

  • Turn distribution agreements into exclusivities.
  • Put the access and integrations your customers depend on under contract.
  • Secure privileged access to the data or platforms your product needs.
03

Service guarantees

What it protects

A service guarantee is the contractual commitment to hold a measurable level of quality, such as availability (SLA), on-call duty, recovery time or insurance. A commit is not enough to produce it. It is operated, measured and proven over time, with teams and procedures behind it. A product rebuilt by an AI reproduces the function, but carries neither the commitment nor the track record that reassure a critical customer.

What I look at to score it

I compare the promise with the measurement. I look for an availability commitment (SLA), then for the monitoring, alerting, on-call rota and already tested backup restore that make it hold. A commitment nobody measures stays at the bottom of the scale.

How to strengthen it

  • Instrument before you promise, with monitoring, alerting and a public status page.
  • Write a recovery plan and test it, backup restore included.
  • Then put availability and on-call duty under contract, at which point they become a commercial asset.
04

Proprietary data

What it protects

Proprietary data is an asset built up over years of operating in the field, made of measurements taken in real conditions that exist nowhere else. An AI can write the code that processes this data, but it cannot invent it. It is often the strongest moat, provided you do not expose it (when a platform closes its APIs, it is often to protect this asset).

What I look at to score it

I look at where the data comes from, measured in the field or taken from open sources, then at who can access it and how it is isolated between customers. Rare data that is exposed, with no governance and no control over retention, is worth less than it looks.

How to strengthen it

  • Isolate data between customers, control its retention and bring it together in a single source of truth.
  • Weigh every API you open against this asset, because opening your data to new entrants exposes your best moat.
05

Sovereignty and reversibility

What it protects

Sovereignty is the right and the technical ability to run the product wherever you decide, on your own premises, at the customer's site or with another cloud provider, without rewriting everything. Reversibility is its counterpart, meaning you can leave a host or a supplier without being locked in. For a sensitive customer (healthcare, public sector, industry), it is a decisive argument, and a piece of architecture work that a quick AI clone almost always neglects.

What I look at to score it

I check whether the product can run somewhere other than its current host. I look at the share of open source in the stack, infrastructure described as code, dependence on a provider's proprietary services and data transfers outside the European Union. Total dependence on a single hyperscaler pulls the score down, even with hosting in a French region.

How to strengthen it

  • Describe the infrastructure as code.
  • Put a portability layer in front of the host's proprietary services.
  • Document an exit strategy, a lever a future acquirer will value too.

Reading the score out of 20

The overall index adds up the five scores. Each score runs from 0, for a moat that is absent or that I do not observe, to 4, for a solid and proven moat. A moat I do not observe still appears on the radar, with a score of 0.

I calibrate the score to the company's stage. In a young startup, a low score describes a path still to build and shows which moats to consolidate first.

The shape of the radar matters as much as the total. The shortest moats are the ones a new entrant equipped with AI will cross first. Prioritising them is risk management, regardless of any transactional consideration. The audit observes the moat, and the investor draws the conclusions.

IRIA remains a quick strategic reading. In the report, I therefore describe the observed state of each moat in the conditional, and keep affirmative statements for facts verified elsewhere in the audit. The index is an expert opinion at a given point in time, based on what I observe. It is neither a certification, nor legal advice, nor a guarantee of valuation.

What the audits show

Across the audits scored so far, the average sits around 7 out of 20. The same findings come back from one audit to the next.

Certifications at rock bottom

It is the weakest moat everywhere. I find no third-party certification, and GDPR compliance stays on paper or partial, even for health, identity or income data.

Declarative service guarantees

The commitment exists in practice, rarely in the contract, never in the measurement. On-call duty, alerting or a tested backup restore are missing.

Data, the best moat yet poorly guarded

Field data is real and hard to rebuild. Yet it stays exposed through leaks between customers, missing governance or uncontrolled retention.

Partnerships without exclusivity

The distribution network or the usage lock-in does exist. None of it is written into an exclusivity contract.

All-or-nothing sovereignty

A portable, open source stack scores very well. Total dependence on a single hyperscaler, often American and with transfers outside the European Union, scores very low.

Technical debt wears the moats down

The technical work streams of the audit (security, monitoring, data modelling, infrastructure as code) also consolidate the moats. Architecture debt directly degrades data and sovereignty.

Frequently asked questions

What is IRIA?

IRIA, the AI Resistance Index, is an index I created and have included in every technical audit since June 2025. It scores from 0 to 4 five moats an AI cannot copy (certifications and standards, locked-in partnerships, service guarantees, proprietary data, sovereignty and reversibility), for an overall index out of 20.

Is IRIA a market standard?

No. It is an analysis framework specific to my audits. No body standardises it, and I never present it as a norm.

Is a low score bad news for a young company?

I calibrate the score to the company's stage. In a young startup, a low score describes a path still to build, and it shows which moats to consolidate first.

Which audits include IRIA?

Since June 2025, all my audits include it, the flash due diligence audit as well as the full technical audit. The report gives it a section right after the executive summary, with the radar of the five moats and a recommendation for each one.

Where do your moats stand?

IRIA is part of every audit I run. A 30-minute call is enough to scope yours.