How This Briefing Works
This report opens with key findings, then maps the gaps between what Cohere discloses and what BLACKOUT observed at runtime. From there: what it means for your organization, what to do about it, and the detection data and evidence underneath.
Key Findings
Subprocessor Disclosure
47 vendors detected on cohere.com, 33 undisclosed
Pre-Consent Tracking
100% pre-consent tracking rate, 17 vendors fire before any consent interaction
Pre-Consent Activity
Cohere was observed loading and executing before user consent was obtained on 100% of sites where it was detected.
Data Anonymization Claims
Deploys 4 B2B deanonymization vendors whose purpose is visitor identification
Undisclosed Party
Not in privacy policy
Claims vs. Observed Behavior
Subprocessor Disclosure
“Trust Center lists 14 subprocessors”
47 vendors detected on cohere.com, 33 undisclosed
Runtime scan shows 6sense, Demandbase, ZoomInfo, MetaPixel, GoogleAds, LinkedIn, Marketo, TrenDemon, HockeyStack and 24 others not in subprocessor list
Pre-Consent Tracking
“Claims to honor GPC and obtain consent”
100% pre-consent tracking rate, 17 vendors fire before any consent interaction
6sense, Demandbase, MetaPixel, GoogleAds, DoubleClick all load pre-consent
Data Anonymization Claims
“Privacy policy claims data is aggregated and de-identified”
Deploys 4 B2B deanonymization vendors whose purpose is visitor identification
6sense, Demandbase, ZoomInfo, RB2B all perform reverse IP lookup and identity resolution
Trust Center Accuracy
“Trust Center represents complete security posture”
Marketing website practices contradict Trust Center claims
SOC2/GDPR/CCPA badges displayed while violating consent requirements
What This Means For You
What To Do About It
Role-specific actions based on observed behavior
If You Use Cohere
- →Audit your Cohere contract against their actual website practices — Trust Center claims do not match runtime behavior
- →Verify their subprocessor list is complete before signing DPA — 33 vendors are missing from disclosures
- →Request explicit confirmation that your AI training data and API usage data are not shared with deanonymization vendors
- →Implement consent-first loading for any Cohere widgets or SDKs on your properties
If You're Evaluating Cohere
- →Be aware that visiting cohere.com exposes your evaluation activity to 6sense, Demandbase, and ZoomInfo before consent
- →Request the complete subprocessor list and compare against runtime detection data before signing
- →Verify Privacy-by-Design claims against BLACKOUT runtime evidence showing 17 pre-consent vendors
- →Negotiate contractual restrictions on competitive intelligence derived from your usage patterns
Negotiation Leverage
- →B2B deanonymization pre-consent: 6sense, Demandbase, and ZoomInfo fire before consent — YOUR AI evaluation activity is visible to competitors; use this to negotiate removal of deanonymization vendors or require consent-first architecture
- →Subprocessor undercount: 47 vendors detected vs. 14 disclosed — 33-vendor gap undermines Trust Center credibility; require complete disclosure as a contract condition
- →Privacy-by-Design contradiction: Marketing claims Privacy-by-Design while firing 17 vendors pre-consent — leverage for enhanced data protection guarantees
- →Enterprise AI sensitivity: AI infrastructure decisions are strategically sensitive — negotiate data isolation and competitive intelligence restrictions for your evaluation and usage data
Runtime Detections
BLACKOUT observed this vendor's JavaScript executing in a live browser and classified each hostile behavior using our BTI-C (Behavioral Threat Intelligence — Capability) taxonomy. These are not theoretical risks — each code below was triggered by something we watched this vendor's code actually do.
Evasion infrastructure, auditor bypass
Keystroke/mouse tracking
Identity stitching
Ignoring CMP signals
Device identification
Container/loader (neutral)
IOC Manifest
Indicators of compromise across 4 categories. Use for detection rules, CSP policies, or Pi-hole blocklists.
Ecosystem & Supply Chain
Evidence Artifacts
Artifacts collected during analysis, available with evidence-tier access.
Complete network capture with all requests and responses
191 detection signatures across scripts, domains, cookies, and network endpoints