VPN Providers — Privacy / FOSS / Anonymity Scoring (Final)

A maximally adversarial scoring pass across five providers: Mullvad, IVPN, cryptostorm, Proton VPN, AirVPN.

As-of date2026-02-28
Scale0–100 per criterion
Compositeweighted sum (100 total weight)
Formatall claims linked inline

Scoring model

Criteria, weights, and grading scale (locked)

Criteria + weights

A. Anonymity by default (signup + payment)
Identity surfaces created during account creation and billing; structural support for anonymous purchase paths.
20
B. Data minimization & logging reality
What is retained by necessity vs. by choice; retention windows; web perimeter logging disclosures.
20
C. Verifiability (FOSS, reproducible builds, independent audits)
Open-source breadth, reproducible artifacts, and public third-party audits.
20
D. Jurisdiction & legal attack surface
Jurisdictional exposure, clarity/ambiguity of corporate posture, and demonstrated minimization under pressure where documented.
12
E. Dependency minimization
Third-party processors, JS gating, analytics/cookies, affiliate/partner attribution economics.
10
F. Protocol & client security engineering
Client hardening and verified security posture; penalties for severe client-side issues with current impact.
10
G. Anti-censorship & advanced routing
Multihop, obfuscation, and operational support for restrictive networks.
8
Composite score = Σ(scoreᵢ × weightᵢ) / 100. All criterion scores are 0–100.

Grading scale

95–100Best-in-class; privacy properties are default and hard to accidentally defeat.
80–94Strong; meaningful tradeoffs exist but remain privacy-first under scrutiny.
60–79Mixed; requires operational discipline or contains structural compromises.
<60Fails this model in that dimension.

High-impact evidence pillars (examples)

Final ranking

Composite + full breakdown (A–G). Provider names link to official sites.

Rank Provider Composite A B C D E F G
1 Mullvad 93.12 98 96 97 78 92 90 92
2 IVPN 88.28 92 90 93 75 88 86 86
3 cryptostorm 78.50 95 82 68 55 70 87 90
4 Proton VPN 77.48 58 92 90 72 58 88 78
5 AirVPN 68.68 78 72 70 62 50 60 78
Final order
1 → 5
Mullvad > IVPN > cryptostorm > Proton VPN > AirVPN
Primary differentiator
Identity
Account lifecycle + payment anonymity surfaces dominate
Secondary differentiator
Verify
Public audits + reproducibility vs. trust-based claims

Provider deep dives

Each section includes score bars, load-bearing wins, hard penalties, and primary sources.

1 Mullvad 93.12composite Anonymous-by-construction + unusually strong verifiability; OpenVPN removed Jan 15, 2026.
Expand
A. Anonymity (20)
98
B. Minimization/logging (20)
96
C. Verifiability (20)
97
D. Jurisdiction (12)
78
E. Dependencies (10)
92
F. Client/protocol security (10)
90
G. Anti-censorship (8)
92

Load-bearing wins

  • Web perimeter minimization: Nginx access logs kept up to 5 minutes (without IPs); no external analytics disclosed in policy — policy.
  • Verifiability: Cure53 relay/infrastructure audit report — PDF, plus audit program write-up — blog.
  • Reproducible builds (Android): starting with version 2025.2 — details.
  • Incentive hygiene: no affiliates/influencers/paid reviews — policy.
  • Anti-censorship tooling: QUIC obfuscation for WireGuard — blog, plus LWO — blog.

Hard penalties (counted)

  • Protocol fallback loss: OpenVPN fully removed Jan 15, 2026 — final reminder.
  • Jurisdiction remains a non-zero legal attack surface in the abstract; mitigation is largely achieved through minimization, not “jurisdictional magic” — minimization disclosures.
Primary sources: no-logging data policy · advertising/affiliates policy · Cure53 (PDF) · QUIC obfuscation · OpenVPN removal.
2 IVPN 88.28composite Extremely strong audits + open source; measurable web telemetry (Matomo JS) keeps it below #1.
Expand
A. Anonymity (20)
92
B. Minimization/logging (20)
90
C. Verifiability (20)
93
D. Jurisdiction (12)
75
E. Dependencies (10)
88
F. Client/protocol security (10)
86
G. Anti-censorship (8)
86

Load-bearing wins

  • Public audit trail: Cure53 web audit report — PDF and IVPN write-up — blog.
  • Apps audit: Cure53 apps + daemon report — PDF and IVPN post — blog.
  • Open source posture: IVPN apps are open source — announcement.
  • Payment posture: accepted methods (including cash and cryptocurrencies) described — payment methods.

Hard penalties (counted)

  • Web telemetry disclosure: Matomo JS collects IP address (then discards last two octets) and may set a cookie — privacy policy.
Primary sources: privacy policy · infra audit write-up · Cure53 2024 (PDF) · Cure53 2022 (PDF).
3 cryptostorm 78.50composite Extreme anonymity primitives + strong obfuscation options; default clearnet web logs cap it.
Expand
A. Anonymity (20)
95
B. Minimization/logging (20)
82
C. Verifiability (20)
68
D. Jurisdiction (12)
55
E. Dependencies (10)
70
F. Client/protocol security (10)
87
G. Anti-censorship (8)
90

Load-bearing wins

  • “Monero order” path explicitly advertises no third parties, no email, no JS required — homepage.
  • Anti-censorship: obfs4 support — blog, plus SSH tunneling — blog.
  • Canary: dated and published — canary.txt.

Hard penalties (counted)

  • Default web perimeter logs: Apache “mostly default logging” includes visitor IP/user-agent/referrer; retained about 2 weeks — privacy page.
  • Jurisdiction clarity is intentionally unconventional; ambiguity counts as attack surface in this model — privacy page discussion.
Primary sources: privacy policy · canary · obfs4 · SSH tunnels.
4 Proton VPN 77.48composite Strong audits + open source; anonymity-by-default is structurally weaker and partner program adds attribution incentives.
Expand
A. Anonymity (20)
58
B. Minimization/logging (20)
92
C. Verifiability (20)
90
D. Jurisdiction (12)
72
E. Dependencies (10)
58
F. Client/protocol security (10)
88
G. Anti-censorship (8)
78

Load-bearing wins

  • No-logs verification: Securitum 2025 report — PDF, and Proton’s audit hub article — blog.
  • Open source + audits: Proton’s “open source and audited” disclosure — blog.

Hard penalties (counted)

  • Bitcoin cannot be used during account sign-up; it can be used after creating the account via credits — Proton doc and Proton VPN doc.
  • Partner/affiliate economics are explicit (“earn up to 100% commission”) — partners program.
  • Legacy OpenVPN manual config retirement has a cutoff date (Feb 28, 2026) for older config files; official apps unaffected — report.
Primary sources: Securitum audit (PDF) · open source + audits · Bitcoin payment constraints · partners program.
5 AirVPN 68.68composite Direct Monero acceptance + FOSS client; JS gating + referral system + severe macOS client CVE drive the drop.
Expand
A. Anonymity (20)
78
B. Minimization/logging (20)
72
C. Verifiability (20)
70
D. Jurisdiction (12)
62
E. Dependencies (10)
50
F. Client/protocol security (10)
60
G. Anti-censorship (8)
78

Load-bearing wins

  • Direct Monero acceptance (“no intermediaries”) is stated on the plans page — buy page.
  • FOSS client repo — Eddie (GitHub).
  • WireGuard privacy caveat is explicitly stated (“by design it is not ideal for privacy”) — FAQ.

Hard penalties (counted)

  • Web dependency: site access in hardened contexts triggers “website require JavaScript” — forum thread.
  • Referral/attribution system exists (“gain money” when another person buys) — referrals FAQ.
  • Client security event: macOS Eddie contains an insecure XPC service enabling local privilege escalation to root (affecting 2.24.6) — NVD CVE; related release context — releases.
Primary sources: plans/payment methods · referrals · WireGuard FAQ · CVE-2025-14979.

Interpretation

What the ordering means inside this model

Why the top two separate

The knife-edge middle

Why #5 drops hard

  • AirVPN keeps some maximalist-friendly properties (direct Monero acceptance: buy page), but dependency and attribution surfaces expand (JS gating: thread; referral system: FAQ).
  • A severe macOS client issue (local privilege escalation to root) is scored as a first-class penalty in “F” (CVE-2025-14979).

Final meaning of the scores

  • #1–#2 are the only providers that score “structurally privacy-first” across account lifecycle + web perimeter + verifiability at once.
  • #3–#4 represent an explicit trade: anonymity primitives vs. third-party verifiability + business incentive surfaces.
  • #5 is suppressed primarily by dependency/attribution expansion and a severe client-side security event.
Return to top.