1.What this is
A dated, verifiable snapshot of what a customer, procurement reviewer or platform auditor can publicly observe about Fernwave's AI transparency posture under EU AI Act Article 50 — with evidence they could check themselves. Every finding carries one of four states:
EVIDENCE FOUND
POSSIBLE GAP
NOT PUBLICLY VERIFIABLE
REQUIRES FACTUAL CONFIRMATION
This is deliberately not a compliance verdict: whether a system is in scope, whether Fernwave acts as provider or deployer, and when systems were placed on the market are factual/legal determinations that cannot be made from public surfaces — where they matter, we say so explicitly instead of guessing.
2.AI surfaces identified (public inventory)
| # | Surface | Where observed | Article 50 relevance |
|---|---|---|---|
| S1 | “Ask Fernwave” in-app assistant (chat) | Product tour, docs, homepage widget | 50(1) — users must know they interact with AI |
| S2 | AI-generated meeting summaries (text output) | Feature page, help center, sample exports | 50(2) — machine-readable marking of synthetic content; 50(4) where summaries are shared onward |
| S3 | Marketing claims about AI (“AI that writes your minutes”) | Homepage, pricing page | Consistency check (claims vs. disclosures) |
| S4 | Underlying model / upstream provider | Not disclosed publicly | Determines provider/deployer role split |
3.Findings
| # | Check | State | What we observed (dated evidence in §4) |
|---|---|---|---|
| F1 | Assistant disclosed as AI at point of interaction (web app) | EVIDENCE FOUND | Chat header reads “Fernwave AI assistant — answers may contain mistakes”; visible before first message. Evidence E1, E2. |
| F2 | Assistant disclosure in mobile app | POSSIBLE GAP | App-store screenshots (current listing) show the chat without any AI label in the visible frames. Only the public listing was reviewed — an in-app label may exist beyond the screenshots. Evidence E3. |
| F3 | Machine-readable marking of generated summaries (50(2)) | NOT PUBLICLY VERIFIABLE | Export formats (PDF/DOCX) are only available to logged-in users; no public sample export found. Cannot be assessed from outside — flagged for internal verification, not assumed absent. |
| F4 | Help-center explanation of AI features and limits | EVIDENCE FOUND | Article “How Fernwave uses AI” (updated 12 Jun 2026) describes the summarisation pipeline and human-review recommendation. Evidence E4. |
| F5 | Provider vs. deployer role for the embedded model | REQUIRES FACTUAL CONFIRMATION | No public statement of the upstream model provider or contractual role split. This determines which Article 50 duties sit with Fernwave — a fact only Fernwave can confirm. |
| F6 | Placed-on-market date relative to 2 Aug 2026 (grandfathering of 50(2) marking until 2 Dec 2026) | REQUIRES FACTUAL CONFIRMATION | Feature launch blog post is dated 14 Mar 2026 (Evidence E5), which suggests pre-existing status, but the legally relevant date is a fact for Fernwave to attest, not for us to infer. |
| F7 | Marketing claims consistent with actual disclosures | EVIDENCE FOUND | Claims are capability-level (“drafts your minutes”), no “human-written” misrepresentation found. Evidence E6. |
4.Evidence log
Each item: captured URL, UTC timestamp, SHA-256 of the stored screenshot/HTML snapshot (synthetic values in this sample). In a real pack the snapshots ship alongside as files; hashes let anyone verify they were not altered after the stated date.
| ID | What | URL (fictional) | Captured (UTC) | SHA-256 (first 16) |
|---|---|---|---|---|
| E1 | Web-app chat header screenshot | app.fernwave.example/tour | 2026-08-04 09:12 | b3a1f09e77c24d10 |
| E2 | Homepage widget, first-open state | fernwave.example | 2026-08-04 09:15 | 4c88d2aa19ef03b7 |
| E3 | App-store listing screenshots (6) | store listing, Fernwave Notes | 2026-08-04 10:02 | 9d41c5f2e8ab7710 |
| E4 | Help article “How Fernwave uses AI” | help.fernwave.example/ai | 2026-08-04 10:20 | 71f0be3c92d4a5e8 |
| E5 | Launch blog post, AI summaries | fernwave.example/blog/ai-summaries | 2026-08-04 10:31 | ea52c7d80f1b3944 |
| E6 | Homepage + pricing claims snapshot | fernwave.example/pricing | 2026-08-04 10:44 | 0fb6a4e19c73d258 |
5.Procurement-ready summary
3
evidence found
1
possible gap
1
not publicly verifiable
2
require confirmation
Attachable answer for questionnaires: “An independent written check of our public surfaces
(Evidale, 4 Aug 2026) found our customer-facing AI assistant disclosed at the point of interaction and our AI
documentation publicly available, with dated evidence on file. Two items are pending internal confirmation
(model-provider role split; marking of exported summaries) and one mobile-surface label is being reviewed.”
That paragraph — honest, dated, specific — is what buyers' reviewers actually want to read. It ages well precisely because it doesn't overclaim.
6.Suggested next steps
- Confirm F5 and F6 internally (10 minutes with whoever owns the vendor contract) — we provide the two attestation templates.
- Close the F2 mobile label gap or document that the in-app state differs from the store screenshots.
- Verify F3 on a real export; if marking is absent, it's a small engineering task with a hard backstop date (2 Dec 2026) for pre-existing systems.
- Re-run the check after your next release — evidence is only as good as its date.
This is a sample. Evidence checks like this one are currently free for a first cohort
of SaaS vendors — reviewed from public surfaces only, delivered in writing in 3–5 business days.
Write to hello@evidale.io.