- Home
- Skills
- Legal & Compliance
- Age Assurance — Estimate, Never Identify
Age Assurance — Estimate, Never Identify
Choose an age-assurance method, compute the challenge threshold from your measured error, and stay inside SREN, GDPR and the AI Act.
$49
Works with the AI Agents you already use
Age Assurance — Estimate, Never Identify
Example session with this skill installed
I need to set up age verification for a French liquor delivery site. We are considering facial age estimation. What is the challenge threshold if the model has a 3-year sigma?
- Read your context and instructions
- Compiled the age assurance
I have calculated the necessary challenge threshold and mapped the legal requirements for your deployment under the loi SREN and ARCOM guidelines.
Age Assurance Strategy: Liquor Delivery (France)
| Metric | Value | Rationale |
|---|---|---|
| Target Age | 18 | Legal minimum for alcohol sales in France. |
| Model Sigma | 3.0 years | Your measured error rate per demographic group. |
| Risk Tolerance | 1% | Maximum allowed probability of a minor passing as an adult. |
| Challenge Threshold | 25 Years | Users estimated below 25 must provide a fallback ID. |
| Friction Estimate | ~92% of 18yo | Percentage of legal adults who will be challenged. |
Regulatory Alignment
Double Anonymity: You must use a third party to verify the "18+" attribute. The site should not know the user's name, and the provider should not know the site visited.
2.
Data Minimization: The selfie used for estimation must be processed in-browser or discarded immediately after the estimate. No images or biometric templates may be stored.
3.
AI Act Compliance: Ensure the vendor does not bundle prohibited biometric categorisation (gender, ethnicity, or emotion recognition) with the age estimate.
4.
Human in the Loop: Since this involves delivery, the final "Passed" status from the app acts as decision support for the delivery driver, who retains final legal responsibility.
Next steps
- Run
python scripts/seuil.pywith your specific demographic bias data to see if the threshold needs to be higher for certain groups. - Define your fallback route (e.g., manual document check) for the ~92% of 18-year-olds who will fail the 25-year threshold.
- Draft the DPIA focusing on why facial estimation is proportional compared to full document storage.
Connects securely to your tools. The creator never sees your data.
About this skill
Estimate an age. Never identify a face. Compute the threshold.
Build age assurance that holds - choose between declaration, verified attribute, document check and facial age estimation, set a challenge threshold from the error you actually measured, test it per demographic group, and keep the design inside French and EU law (loi SREN, ARCOM, GDPR, AI Act). Use when a site must restrict access to adults, a venue checks ages at the door, or an age-estimation vendor must be assessed before being trusted.
Two ideas carry the skill. Estimating an age is not recognising a face: the estimator reads an image, returns a number, and the image is gone - no template, no database, no matching, no identity. And you never deploy "is this person 18?" - you deploy a challenge threshold, below which the person is asked for something stronger. Treating the estimate as a verdict lets minors through and insults adults, in proportions nobody measured.
Includes an EXPERT.md reference (exact parameters, thresholds, failure diagnosis): the threshold mathematics, the MAE-to-sigma conversion vendors never do for you, a worked friction table, the per-group testing protocol, the legal map (loi SREN 2024, ARCOM référentiel and double anonymity, GDPR art. 9 and 35, AI Act art. 5, UK Online Safety Act), a ten-point vendor assessment checklist and the failure table.
Tested scripts: scripts/seuil.py - it turns a measured error into a defensible threshold and shows what it costs. With a model at sigma 3 years, protecting a 16-year-old at 1 % risk, the threshold is 23 years, and 95 % of 18-year-olds and 74 % of 21-year-olds get challenged. It refuses when the model is too imprecise, when the threshold differs too much between demographic groups, or when only one group was measured.
Acceptable use: restricting access to age-restricted content, goods, venues and services where the law requires or allows it, with the person informed and a non-biometric alternative always available. The skill refuses facial recognition, identification or matching, storing or re-using face images, biometric categorisation by ethnicity, gender or emotion, covert estimation, estimating minors' ages for profiling or advertising, and using an estimate as sole proof with no fallback.
Not legal advice. The references are a starting point, the texts change, and a deployment should be reviewed before it goes live.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Legal & Compliance.
Does not process images or perform biometric analysis. Requires external error rate measurements (sigma) to calculate thresholds. Scope limited to French and EU regulatory frameworks.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 27 days ago
- Software
Python