benchmark · Jul 28, 2026

AIRIN Quarterly Benchmark · Q2 2026

State of AI Platform Data Rights

Rating changesWhat moved, and why it matters.

Risk by categoryTraining, sharing, retention, ownership.

Verbatim proofEvery claim points back to source text.

This is the AIRIN paid-report direction: a visual, board-ready benchmark report backed by verified policy evidence. Informational only, not legal advice.

The market in one line

The quarter's defining signal

The market is moving toward clearer training controls, but the broadest clauses still live in vendor development, improvement, and data-sharing language.

3

sample platforms

0

HIGH-risk findings

303

verbatim citations

3x

weekly re-scans

Our Usage Policy (also referred to as our “Acceptable Use Policy” or “AUP”) applies to anyone who can submit inputs to Anthropic’s products and/or services, including via any authorized resellers or passthrough access, all of whom we refer to as “users.” The...

Top cited sample: Claude (Anthropic) · Usage Policy › “Usage Policy \ Anthropic”

Risk distribution

Platform risk ratings, sample stack

HIGH

0

MEDIUM

105

LOW

198

Band	Findings	Meaning
HIGH	0	Read before adopting or renewing.
MEDIUM	105	Usable with controls and clear owner.
LOW	198	Protective or lower-friction language.

Policy shifts · Q2 2026

Four changes buyers should notice

MEDIUM -> HIGH

Claude (Anthropic): data sharing

Changed from third_party_or_vendor_sharing to sale_or_sell.

HIGH -> MEDIUM

Claude (Anthropic): data sharing

Changed from sale_or_sell to third_party_or_vendor_sharing.

HIGH -> MEDIUM

Claude (Anthropic): model training

Changed from training_permitted to training_with_opt_out.

MEDIUM -> HIGH

Claude (Anthropic): model training

Changed from training_with_opt_out to training_permitted.

Exposure surface analysis

Where the risk actually lives

moderation enforcement

MEDIUM

0 high, 24 medium, 63 low findings in this sample.

privacy data use

MEDIUM

0 high, 18 medium, 85 low findings in this sample.

subprocessors data sharing

MEDIUM

0 high, 17 medium, 6 low findings in this sample.

The useful report does not just say “risky.” It names the policy surface, shows the clause, explains the rating, and keeps the proof attached.

Methodology

How AIRIN proves a report

1

Verbatim capture

Policy text is captured from source documents and stored with exact quotations.

2

Cryptographic hashing

Snapshots keep a SHA-256 trail so the evidence can be re-opened later.

3

Deterministic rubric

The same clause should produce the same risk rating every time.

4

Silence is caution

If a vendor does not clearly protect a right, the report says so plainly.

5

Dual-lens review

Creator, GRC, and counsel views are separated so one audience does not flatten another.

Full report

What the complete paid report includes

30-40 page visual PDF report
Quarterly rating changes and why they moved
All HIGH-rated platforms analyzed
Five biggest policy changes with verbatim quotes
Industry trend analysis
Platform rankings by category
Legal, procurement, and board-deck-ready excerpts
Proof footer with citations and methodology

Quarterly Benchmark Report

Built for people who need a fast, serious answer with the evidence still attached.

$449