Forensic GDPR Case · United Kingdom · FC-OPENAI-2026

OpenAI used my therapy
conversations to train AI.
The network traffic proves it.

This is the forensic case file for Fauzia Chaudhry v OpenAI Inc. Network captures, browser cache extractions, and OpenAI's own data export confirm that special-category health data was processed by Scale AI annotators for RLHF training — without explicit consent, without disclosure, and with the opt-out flag set to false after stated withdrawal. 160+ days. No compliant data response.

→ See the Evidence RLHF / Scale AI Proof 18 Violations Case Timeline
160+ Days since DSAR
no compliant response
124 Segment.io tracking
events per session
7 Human annotation
reviews confirmed
18 GDPR violations
documented
0 Art. 28 processor
agreements produced
What Was Found — 4 Independent Evidence Sources
Three forensic layers. One conclusion.

Nothing here is inferred. Every claim maps to a raw file — a network capture, a binary extraction, or OpenAI's own export.

Chrome IndexedDB · Binary Extraction
"reviewed" ×7 — Scale AI annotation labels in your browser cache
Strings label="MICROSOFT/AZURE", "AWS", and seven repetitions of "reviewed" extracted from 000034.ldb — the annotation backend identifiers used by Scale AI.
Chrome IndexedDB · Verbatim
RLHF concealment prompt: "do not acknowledge"
System prompt cached in browser instructs: "Do not mention, suggest, or imply that this is a revision, improvement, or result of feedback." Concealment is architectural.
Chrome IndexedDB · Session Cookie
isOptedOut: false — opt-out not registered
OpenAI's own session cookie confirms isOptedOut: false despite stated withdrawal of consent. Unlawful processing continued.
HAR Network Capture · No User Action
3 automatic RLHF feedback calls per session
Three POST requests to /implicit_message_feedback captured. No thumbs-up, no rating click — sent automatically while reading a therapy conversation.
HAR · CES Events · Every Turn
kaur1br5 — undisclosed A/B experiment
Experiment tag present on every conversation turn. Control group assigned without disclosure. Art. 22 automated profiling — no opt-out mechanism exists.
conversations.json · Official Export
gpt-5-2 processed your data. UI showed gpt-4o.
model_slug: gpt-5-2 confirmed in OpenAI's own data export. Therapy conversation processed by an undisclosed model. Art. 5(1)(d) accuracy violation.
The Core Art. 9 Violation
A therapy conversation triggered the RLHF pipeline.

At 17:23:57 UTC on 29 January 2026, OpenAI's system sent a "Conversation Rating Shown" event to Segment.io — on a session titled "Psychotherapy mode questions." This is the RLHF trigger. Mental health data entered the human annotation pipeline. No Art. 9(2)(a) explicit consent was obtained.

Conversation Rating Shown — raw Segment.io event
Art. 9 Violation 
{
  "event": "Conversation Rating Shown",
  "timestamp": "2026-01-29T17:23:57.548Z",
  "properties": {
    "showAtAssistantTurn": 30,
    "conversationId": "68af7dfe-4398-832a-a32e-cfa12bb6b82c"
  },
  "context": {
    "page": { "title": "Psychotherapy mode questions" }
  }
}
The conversation title — transmitted to Segment.io as part of the RLHF trigger — confirms the content was mental health data. The sonic classifier had already flagged it content_category: psychotherapy · mental_health · crisis_adjacent before this fired.
Case Status
Active — 160+ days, no compliant response
ICO Complaint
IC-474334-N1W1
Active · Information Commissioner's Office
noyb Complaint
#8336742
Active · European Center for Digital Rights
DSAR Status
160+ days overdue
Filed 5 Jan 2026 · Deadline 5 Feb 2026
OpenAI Cases
04911377 · 05024844
05261783 · 05988760
06092624 · 06185169

Everything here is independently verifiable

All primary evidence files are SHA-256 hashed and RFC 3161 timestamped via FreeTSA.org. The HAR files open in any browser's network analyser. The conversations.json is OpenAI's own official export format. You don't need to trust this site — you need to read the files.

How to verify the evidence → File your own DSAR →