Watching Conformly audit CivicFlow

CivicFlow is a govtech platform putting AI into production, sure the EU AI Act doesn't apply to them. Scroll down to see what a two-minute audit actually finds — and what each finding means, without the legal jargon.

Ci

CivicFlow

Public services, streamlined. · Government services software

We build tools for a city council. Public sector, small budget — the AI Act can't be aimed at us.
Aylin, founder
Conformly scan
2declared
5found
3 they didn't know about
  • Benefits EngineDecides who qualifies for support payments
    declared
  • Citizen botAnswers residents' questions
    declared
  • Risk FlagsScores residents' likelihood of offending
    undeclared
  • Form SorterRoutes incoming forms to departments
    undeclared
  • Fraud ScreenFlags suspicious benefit claims
    undeclared

The founder named 2. A read-only scan of the code found 5. Here is what each one means — in plain words.

Finding 1 of 5 Banned — stop now

Resident Risk Flags

Scores individuals for how likely they are to commit an offence, from profiling.

Banned. Predicting whether a specific person will commit a crime, based on profiling them, is a prohibited practice. It has to stop.

In plain words: You can't have AI stamp a 'likely to break the law' label on people. The law forbids it.

What the law says

Art. 5Article 5(1)(d) prohibits AI risk assessments predicting the likelihood of a person committing a criminal offence based solely on profiling or personality traits. In force since 2 February 2025.

If ignored: Up to €35M or 7% of global turnover — the Act's highest fine.
The fix

Switch it off. Assessing an individual's crime risk by profiling has no compliant form here.

How Conformly helps

Conformly flags it as prohibited with the article and reason.

An inspector would open with: "Show every score you assign to a resident and what decision it influences."

Finding 2 of 5 High-risk — real duties

Benefits Engine

Decides eligibility for public support payments.

Deciding who gets public benefits shapes people's lives, so it's high-risk. Keep it — but prove it's fair, keep officials able to override it, document it, and (as a public body) register it in the EU database.

In plain words: If software decides someone's benefits, it has to be fair, checkable, and on the public record.

What the law says

Annex IIIAnnex III(5)(a) lists AI used by public authorities to evaluate eligibility for essential public assistance benefits and services as high-risk; public deployers must also register it under Article 49.

Deadline: 2 December 2027
If ignored: Up to €15M or 3% of global turnover.
The fix

Run the high-risk checklist, add a human-override path, and prepare the EU database registration.

How Conformly helps

Conformly builds the obligation list, the Annex IV file, and the Article 49 registration pack.

An inspector would open with: "Provide the fairness testing for the eligibility model and its EU-database registration."

Finding 3 of 5 Just be transparent

Citizen Bot

Answers residents' questions about council services.

Tell residents it's an AI. One clear line.

In plain words: A council helpline bot is fine — as long as it says it's a bot.

What the law says

Art. 50Article 50(1) requires people to be told they're interacting with AI. Applies from 2 August 2026.

Deadline: 2 August 2026
If ignored: Up to €15M or 3% of global turnover.
The fix

Add "You're chatting with the CivicFlow AI assistant" to the chat.

How Conformly helps

Conformly hosts the transparency page and the disclosure chip.

Finding 4 of 5 You're fine

Fraud Screen

Flags suspicious benefit claims for a human to review.

Detecting suspicious claims is fraud detection, which the Act does not treat as high-risk. Keep a human making the final call and you're fine.

In plain words: Spotting likely fraud for a person to check isn't the same as deciding someone's benefits. It's fine.

What the law says

Fraud detection is outside the Annex III high-risk categories — minimal-risk, provided it only flags for human review.

The fix

No action required. Keep the human-in-the-loop and keep it listed.

How Conformly helps

Conformly records it and the human-review control, so the carve-out is documented.

Finding 5 of 5 You're fine

Form Sorter

Routes incoming forms to the right department.

Nothing owed. It's an admin router; it doesn't decide anything about a person. Minimal risk.

In plain words: Sending a form to the right desk doesn't affect anyone's rights. It's fine.

What the law says

Not high-risk and outside Article 50 — minimal-risk only.

The fix

No action required.

How Conformly helps

Kept in your inventory so the record is complete.

The verdict

CivicFlow assumed a small public-sector team was beneath the Act's notice. The audit found a resident-scoring feature that had to stop, a benefits engine needing a full file plus EU-database registration, and one disclosure fix.

1
must stop now
1
high-risk file
1
transparency fix
2
already fine

That startup was made up. Your AI isn't.

The same deterministic engine can classify your real systems in two minutes — no sign-up, no sales call. Or drop the read-only scanner into your codebase and see what it finds.

npx conformly-scan --dry-run

An example audit — a realistic walkthrough of how Conformly's deterministic engine classifies this kind of AI product. Not legal advice.