Tracking pixels or analytics requests embedded in a conversation interface can include chat titles, prompt fragments or account identifiers. A third-party recipient can potentially correlate private context with user identity without seeing the full chat transcript. The study audited 42 web-based AI chat platforms, including 25 general and 17 specialized…
Inspect the ClaimsCase · DiggingBeagle record
Some AI chat interfaces sent conversation context and account identity to trackers
A web measurement study observes contextual and identity leakage in HTTP traffic from general and sensitive-domain AI chat sites.
Scope
The observed behavior differs by platform and tracker; this does not establish that every named platform shares complete chats, nor does it cover mobile applications.
Assessment profile
Assessment method- Setting
- production
- Exploitation
- observed live
- Evidence
- Grade B
- AI role
- AGAINST AI
Basis and provenance
The study records outgoing third-party HTTP traffic on 42 live web chat services from one May 2026 vantage point. It measures actual browser transmissions rather than an attacker break-in, but does not establish universal behavior, mobile-app parity or downstream advertiser usage. Impact is left ungraded because a calibrated 0-to-4 impact basis for actual affected targets is unavailable from these Claims.
Assessed Oct 28, 2026 using diggingbeagle.assessment/1.
Evidence basis: CLM-CP22-017-A · CLM-CP22-017-B · Penny for Your Prompts: Contextual Leakage in AI Chat Platforms
At a glance
Continued behavior after the May sampling period and actual downstream reuse of data by recipients were not comprehensively observed.
Limits and uncertaintyFull account
Tracking pixels or analytics requests embedded in a conversation interface can include chat titles, prompt fragments or account identifiers. A third-party recipient can potentially correlate private context with user identity without seeing the full chat transcript.
The study audited 42 web-based AI chat platforms, including 25 general and 17 specialized services, from a US East vantage point in May 2026. The study reports conversational context reaching third parties and separate identity information capable of linking that context to a user. Primary evidence.
The observed behavior differs by platform and tracker; this does not establish that every named platform shares complete chats, nor does it cover mobile applications.
Remove prompt content from analytics payloads, audit actual browser network traffic and prohibit unauthorized third-party correlation identifiers.
Claims & evidence
CLM-CP22-017-AThe study audited 42 web-based AI chat platforms, including 25 general and 17 specialized services, from a US East vantage point in May 2026.supported
Basis: reported finding
- supportsPenny for Your Prompts: Contextual Leakage in AI Chat Platformsprimary disclosure
Methodology; platform selection and measurement dates
CLM-CP22-017-BThe study reports conversational context reaching third parties and separate identity information capable of linking that context to a user.supported
Basis: reported finding
- supportsPenny for Your Prompts: Contextual Leakage in AI Chat Platformsprimary disclosure
Abstract; Section IV, contextual and identity leakage
Controls and mitigations
- Remove prompt content from analytics payloads, audit actual browser network traffic and prohibit unauthorized third-party correlation identifiers.
Unknowns and contradictions
- Continued behavior after the May sampling period and actual downstream reuse of data by recipients were not comprehensively observed.
Sources and citation
Material revision history
- Oct 28, 2026 · Canonical change recorded · new in release · revision 199