INDEPENDENT COMPLIANCE AUDITING

Does your site or app pass?

Reviews for online shops, websites and AI-built applications. We identify potential problems, record evidence and distinguish what was confirmed, what needs fixing and what has not been tested.

A site address runs a check on the live page. A public GitHub repository returns a sample of the code. Both are free.

Do you sell from this site?

Sites only: consumer-law criteria apply only if you sell.

See what the free code sample covers →

  • Free
  • No account
  • Under a minute
28/06/2025 Accessibility mandatory for online shops · fines up to €100,000 · Greek Law 4994/2022 45% of AI-written code contains a security flaw · Veracode 2025 19/06/2026 Withdrawal needs a working control, not just text · Dir. 2011/83 art. 11a 96% of developers do not fully trust AI-written code · Sonar 2025/26 20/07/2025 The EU ODR platform shut down · Reg. (EU) 2024/3228 48% of developers review the code their AI writes · Sonar 2025/26 2,000+ security flaws across 5,600 public AI applications · Escape.tech 2025 27/09/2026 Harmonised legal guarantee notice becomes mandatory · Reg. (EU) 2025/1960

Two cases, one standard

FOR ONLINE SHOPS

Deadlines that have already passed

One-click withdrawal, accessibility, trader identity, a link to a platform that shut down. These are not our opinion — they are dates in EU law, and most of them are behind us.

What we check on a shop →
FOR APPLICATIONS

Built fast, often with AI

An application now takes weeks to build. Whether it survives an attack, a payment edge case or a GDPR audit is a different question — and one almost nobody has asked.

What we check on an application →

Why it matters

Two different risks, one cause: nobody checks. For shops the law already has deadlines that have passed; for applications, code ships without a second pair of eyes.

For shops: dates, not estimates

  • 28/06/2025 Accessibility became mandatory for most online shops (European Accessibility Act, Directive (EU) 2019/882). Fines reach €100,000 in Greece.
  • 19/06/2026 Withdrawal must be a function, not just a paragraph (Directive 2011/83/EU, Article 11a).
  • 20/07/2025 The EU ODR platform shut down. Anyone still linking to it is sending customers to a page that no longer exists.

For applications: what the research shows

45%

of AI-written code contains a security flaw.

Source: Veracode, GenAI Code Security Report 2025

48%

of developers review AI-written code, while 96% do not fully trust it.

Source: Sonar, State of Code 2025/26

2,000+

security flaws and 175 personal-data leaks across 5,600 public AI-built applications.

Source: Escape.tech, 2025 research

Who it is for

Online shop owners

You sell online and you do not know whether your site meets what the law asks today. The audit answers with dates, not opinions.

People who build and sell

Startups, software houses and founders who built an application and want to prove to customers that it is serious, before it reaches the market.

People who buy or invest

Investors, acquirers, procurement teams. An independent view of what they are buying, before they sign.

What we check: the whole path, not just security

Most audits look at security alone. We look at the whole path — from what the site promises to what it does in production — against international standards (OWASP ASVS 5.0, ISO/IEC 25010, WCAG 2.2, NIST SSDF) and EU law.

GATE G0

Product

Who it serves, which problem it solves, which risk class it falls into.

Risk class

Audit path

G0 · Product G9 · Consumer rights

GATE G1

Requirements

Written features with acceptance criteria, a personal-data inventory, a threat analysis.

Acceptance criteria Threat model

Audit path

G0 · Product G9 · Consumer rights

GATE G2

Architecture

Tenant separation, roles and permissions, which laws apply, DPIA.

Roles DPIA

Audit path

G0 · Product G9 · Consumer rights

GATE G3

Code quality

Static analysis, duplication, hallucinated libraries, silently swallowed errors — the typical failures of AI-written code.

Static analysis

Audit path

G0 · Product G9 · Consumer rights

GATE G4

Security

OWASP ASVS, secrets in code, vulnerable packages, whether one customer can see another’s data.

OWASP ASVS

Audit path

G0 · Product G9 · Consumer rights

GATE G5

Testing

Whether the tests catch real defects, plus mobile, speed and load.

Tests Speed

Audit path

G0 · Product G9 · Consumer rights

GATE G6

Operations

Backups with a tested restore, monitoring, rollback, incident process.

Backup Rollback

Audit path

G0 · Product G9 · Consumer rights

GATE G7

Legal

Terms, GDPR and DPAs, cookies, the AI Act, software licences.

GDPR AI Act

Audit path

G0 · Product G9 · Consumer rights

GATE G8

Accessi­bility

WCAG 2.2, touch targets, contrast, keyboard navigation — measured on the live page.

WCAG 2.2

Audit path

G0 · Product G9 · Consumer rights

GATE G9

Consumer rights

Withdrawal, deadlines, trader identity, prices and charges — as Directive 2011/83/EU requires.

Dir. 2011/83

Audit path

G0 · Product G9 · Consumer rights

How the audit works

Full methodology →
  1. Connect

    For a shop, the address is enough. For an application, read-only access to the repositories you choose. The audit runs in our own ephemeral environment — not on your infrastructure, so that it stays independent.

  2. Tooling

    The audit runs in an isolated environment inside the EU. Security flaws, secrets, packages, tests, speed, accessibility, consumer rights.

  3. Three AI models, blind

    Claude, GPT and Gemini read separately, with no sight of each other. Then each one reviews the others’ findings.

  4. A human

    Where the models disagree, on every serious finding, and on anything high-risk, a senior engineer decides.

  5. Result

    A report with findings and fixes. You fix, you resubmit. On a pass: grade, certificate, badge.

Why three models and not one? On our first real audit, the second independent reviewer found 3 serious flaws the first had missed — among them one that allowed signing into somebody else’s account. One reviewer, however good, is not enough.

What you get

  • A report of findings

    Every finding with file and line, why it matters and how it is fixed.

  • A priority list

    What to fix now, what before your first customer, what later.

  • A signed certificate

    Digitally signed to the W3C standard, recorded in a public registry.

  • A live badge for your site

    Always shows the current state. If it expires or is revoked, it changes immediately.

  • A Wallet card

    The certificate on your phone, with a QR code for verification at meetings and trade shows.

  • Continuous re-checking

    The fast criteria are re-checked on every change, not once a year.

codeauditedSILVER · ACTIVE
APPLICATION
Acme App
GRADE
A
ISSUED
09/2026
EXPIRES
03/2027
Scan for
live verification

The badge

This is what goes on your site when you pass. It is not an image we send you — it is drawn by us every time somebody looks at it.

The badge, in its three forms
The seal is built for a light background — here it sits the way it will sit on your own site.

Silver · ACTIVE

codeauditedSILVER · ACTIVE
APPLICATION
Acme App
GRADE
A
ISSUED
09/2026
EXPIRES
03/2027
Scan for
live verification

VERIFICATION

W3C signature
valid
Public record
logged
Status
ACTIVE
Level
Silver

The seal shows an active certificate.

  • It cannot lie

    If the certificate expires or is revoked, the badge shows it at that moment, inside your own site, with nobody going in to change it.

  • It clicks through

    Straight to the public report: what was audited, when, at which commit, with what result. Anyone who doubts it checks in ten seconds.

  • It states which audit

    Bronze, Silver or Gold is written on it. Nobody can display a badge for a deeper audit than the one they passed.

Why trust us

A mark is worth exactly as much as the discipline behind it. These are our commitments, and they are publicly checkable.

  • We do not sell fixes

    Whoever certifies must not profit from finding problems. The fixes are yours to make, or whoever you choose.

  • Published criteria

    The standard and every threshold are public. Nobody passes on a discount.

  • Revocations are public

    If a site degrades after certification, the badge comes down and the registry shows it.

  • Separation of roles

    Whoever wrote the code does not certify it, and the reviewer has no commercial relationship with the client under audit. The conflict-of-interest rules are enforced by the system itself.

  • Measured accuracy

    We measure our method against a test corpus with known answers, and we publish the result when the measurement is complete. Until then we quote no percentage.

  • Appeals

    Disagree with a finding? There is a written appeals process with a response deadline.

  • Money-back guarantee

    If we miss a serious flaw, we refund the cost of the audit and repeat it.

    See the guarantee →
Independence policy →

What we found

We ran the audit against public repositories holding AI-built applications, and against our own. These are the numbers, as they came out.

19 applications audited
9.693 findings in total
52 critical findings
15 applications with at least one critical

The repositories are not named. Their code is public and the audit lawful, but we notified no author and nothing was fixed. Publishing a vulnerability by name, without warning, is not an auditor’s job.

Measured on 23 October 2026. The totals are produced from the audits themselves by command, not written by hand.

And our own

  • Dradera 1 critical · 5 high fixed
  • Command Center 2 CVE σε εξαρτήσεις open
  • Thelimata 178 αρχεία · 0 ευρήματα clean

Auditing our own applications also found a fault in the auditor itself: it stopped on every Laravel application with a storage symlink. Fixed the same day. We write it here because a scheme that hides its own faults has nothing to say about anyone else’s.

Four packages. A certificate and a badge are issued ONLY with three independent models — with one or two you get a report, not a certification.

Audit with 1 model

€290

Every automated tool and one AI model across the whole codebase. Findings report. No certificate.

Audit with 2 models

€590

Two independent models from different companies, cross-checking each other. No certificate.

Recommended

Bronze — 3 independent models

€1,200

Three models from three companies. The full method, so a certificate and a badge.

Silver — 3 models + engineer

€3,500

Everything in Bronze, with a senior engineer on every serious finding and a named auditor signature.

See all levels and prices

Frequently asked questions

What exactly does the free check see?
Only what is visible on your public page, like any visitor. It does not probe for vulnerabilities, fill in forms or place orders, and it respects your robots.txt.
Do you see our code? Do you keep it?
On an application audit, read-only. The code is deleted as soon as the findings are written, and within 30 days in any case. We keep the findings and the audit trail — not the code.
Are the AI models trained on our code?
No. We use the professional APIs under terms that forbid training, with processing inside the EU.
Are you a government or accredited body?
No. We are a private, independent mark with published criteria. We do not replace obligations set by law.
What if someone tries to fool the AI?
The audit is designed so that no text inside the code can influence the result, and any attempt fails the audit automatically.
How long does it take?
The free check finishes in under a minute. The automated full audit completes the same day. Human review on Silver and Gold depends on reviewer availability, which we state in the quote.
Do you guarantee it cannot be breached?
No, and nobody can. We certify that at the moment of the audit it met the published criteria, and we keep re-checking.
What if you miss something?
We refund the cost of the audit and repeat it free of charge. On Gold we also pay a bounty to whoever found it.

Find out where you stand.

The free check shows the serious ones before somebody else finds them — a customer, a competitor or a regulator.

Start the free check