How we operate, and what you can hold us to.

Independence

AI Analytics is operator-led and self-directed. We choose our targets, we choose our collaborators, and we publish what we've verified — regardless of who is asking. We do not accept work that conflicts with the open-data mission of Voidly. We do not place stories. We do not accept editorial input from funders, partners, or contracted customers on the censorship dataset.

Open datasets and proprietary software have distinct access and licensing terms. Swarm SDK's public materials describe its architecture; its implementation is proprietary. The project status states the limits of what the public materials establish.

Access, licensing & independence

• Published terms. Each public dataset states its compilation license and underlying source terms. Open-data licenses do not grant access to proprietary application code. See data access and the relevant project documentation.
• No grant-conditional work. Where research grants are accepted, the dataset and methodology remain open and uncoupled from the funder's policy preferences.
• No advertising, no surveillance monetization. The site does not run trackers, ad networks, or session-replay tools.

Ethics & method

• Verifiable claims. Every public number cites a source — a live counter at voidly.ai, a GitHub README, or the linked writing post.
• Cross-source verification. Censorship incidents are promoted to "verified" only after correlation against at least one independent project (OONI, CensoredPlanet, or IODA).
• No participation in offensive operations against civilian infrastructure, journalists, or human-rights defenders.

Data handling

• Voidly measurement data. Probe results are published in aggregate under CC BY 4.0. No personally identifying information is collected from probe operators or end users.
• Site analytics. No first-party analytics. No cookies set by application code. Cloudflare error reporting (NEL) is enabled at the edge for service health; no behavioral tracking.
• Inquiry data. Email content is retained only as long as needed to respond. Arrange a suitable channel before sending sensitive material.
• Warrant canary. 0 warrants received as of the 2026-06-22 update above; unsigned pending key rotation. 0 as of last publication.

Conflict of interest

We disclose potential conflicts proactively. Where a writing post discusses a vendor or platform with which we have a commercial relationship, that relationship is stated at the top of the post. Where a probe is hosted by a sponsor (e.g., a research lab donating edge capacity), the sponsor is named on the methodology page.

Response SLAs

General inquiries

24–48 hours

Security disclosures

24 hours · acknowledgment

Press / journalist verification

24 hours business · 48h weekend

Government partner queries

Same business day

Security disclosure flow

1. Send a short initial report without sensitive material. Arrange a suitable channel for any additional details.
2. Email info@ai-analytics.org with a clear subject line and reproducible steps.
3. Acknowledgment within 24 hours. Triage and remediation timeline within 5 business days.
4. Coordinated disclosure preferred. We will credit reporters in the security acknowledgments page on request.
5. Full policy at /.well-known/security.txt (RFC 9116).

Corrections, right of reply, takedown

We index 50M+ source rows across datasets from government sources, regulators/SROs, and derived indexes. That total is not a count of distinct pages or URLs. Source publishers retain their own corrections cadences, and you can also request corrections, right-of-reply annotations, and takedown review directly:

• Corrections & right of reply — fix transcription errors, add disposition notes for dismissed cases, attach right-of-reply statements (48–72 hour response).
• Takedown & legal process — DMCA designated agent, subpoena / court-order intake, preservation requests, abuse reports.