Security audits explained: what 'audited' does and doesn't tell you
An independent security audit is a time-boxed check of one defined part of a product. It is a useful signal, not a guarantee, and "audited" rarely says which part was checked.

Security audits explained: what “audited” does and doesn’t tell you
An independent security audit is a time-boxed check of one defined part of a product. It is a useful signal, not a guarantee, and “audited” rarely says which part was checked.
Who this guide is for
Anyone who picks a VPN, messenger, password manager or notes app partly because it is “independently audited”, and wants to know how much weight that word deserves. It is most useful before you trust a product with something sensitive, and when two products look equal except for their audit claims.
What this guide does not solve
- It does not rate products. The reviews do that; this page explains how to read the audit line inside them.
- It is not a list of audited products. Audits go out of date quickly, so a list would too.
- It cannot tell you a product is safe. No outside check can. It helps you judge how much an audit tells you.
What an audit actually is
Take a real report. In 2024 the security firm Cure53 tested the relay servers of Mullvad VPN. Their public report says:
- Mullvad approached Cure53 in January 2024 and the test ran in June 2024.
- A team of three senior testers was assigned, and the test took sixteen days in total.
- It was a white-box test: Mullvad handed over the source code and the access the testers needed.
- The scope was one staging WireGuard relay, one staging OpenVPN relay, and the repository with the configuration used to deploy them.
- Result: five findings, two of them classed as vulnerabilities (one low, one medium) and three as general weaknesses. Cure53 called the system “well-designed, well set up, and generally in exemplary condition.”
That is a good result, and it is also narrow. The report says nothing about the Mullvad app, about other parts of the service, or about the production servers as they run today. It describes one set of systems over sixteen days in mid-2024. Every audit has this shape: a scope, a number of days, a version or date, and a list of findings.
Not all “audits” are the same kind of check
| Kind | Example | What it tells you | What it does not tell you |
|---|---|---|---|
| Code and penetration audit | Cure53 on Mullvad’s relays (2024) | Testers looked for exploitable bugs in the named components | Anything outside the scope, or anything changed since |
| Cryptographic protocol review | ETH Zurich researchers on Threema (2022-23) | Whether the encryption design itself holds up | How the apps are coded, or how the company runs its servers |
| No-logs assurance engagement | Deloitte Audit Lithuania for NordVPN (2024) | Auditors interviewed staff and inspected servers and technical logs against the company’s no-logs claim, for an assessment window (access ran 18 November to 20 December 2024) | How the company behaves outside that window, or what the code does. NordVPN publishes a summary; the full report is available after logging in to a Nord Account |
| Compliance report or certificate | SOC 2 (AICPA); ISO/IEC 27001 | Whether a company runs documented security processes. AICPA describes SOC 2 as an examination of a service organisation’s controls relevant to security, availability, processing integrity, confidentiality or privacy | Whether the product’s code is secure. For ISO 27001, ISO itself does not certify anyone: outside certification bodies do |
The word “audited” is used for all four, and they answer different questions. A no-logs assurance report tells you about a company’s systems and procedures. It does not tell you the VPN app is free of bugs. A code audit tells you about bugs. It does not tell you what a company does with your data.
When “audited” missed something
Threema is a useful case because it was audited and still had flaws found later. Before 2022, Threema’s audit list shows two audits: 2019 (a university lab) and 2020 (Cure53, covering the apps). Cure53’s 2020 audit took sixteen person-days and found no serious vulnerabilities. In October 2022, researchers at ETH Zurich told Threema about seven attacks against its protocol, and published them in January 2023. They noted that previous independent audits of Threema “did not review the cryptographic core.”
Threema’s side: in a public statement it said the findings had been addressed or no longer applied to its current protocol, that none ever had considerable real-world impact, and that its new protocol, Ibex, had been planned for some time. The researchers record mitigations in phases between Q4 2022 and Q1 2023, with Ibex introduced in November 2022.
Both things can be true. The earlier audits were probably done well for what they covered. What they covered simply did not include the part that the later research attacked. Threema’s own 2020 audit page says “the Threema apps” and does not spell out whether the protocol design was in scope. That is the lesson: a clean audit of the apps is not a review of the encryption design, and the word “audited” hides the difference. Threema has since added a 2023 analysis of its new protocol by university researchers to its audit list. The Threema guide covers the product itself.
How to read an “audited” claim
Six questions, in the order they are quickest to answer:
- What exactly was audited? The app, the servers, the encryption design, or the company’s procedures? Match it to the claim. A no-logs claim needs a server and log check, an encryption claim needs a protocol review.
- How old is it, and which version? An audit of last year’s code says little about this year’s release. Repeat audits, like Mullvad’s across several years, are a stronger signal than one old report.
- Is the report public? A full PDF with findings is worth more than a blog post. A summary tells you the company is comfortable with its own summary.
- How big was it? The report usually states days and testers. Sixteen days by three testers is a serious look at a small scope, not a review of everything.
- What did it find, and was it fixed? An audit with zero findings is less informative than one that lists issues and says what changed. Mullvad’s report calls a small number of findings “a positive sign” and recommends fixing them quickly; look for a follow-up that confirms it.
- Who commissioned it? Mullvad, NordVPN and Threema each hired the auditor in the examples above. That is normal and does not make the result false, but it usually means the company also helps define the scope. The ETH Zurich research was not commissioned by Threema, and it looked at the part the earlier audits had not reviewed.
When a missing audit matters
No published audit is not proof of a problem. Plenty of well-run open-source tools have none. Our own reviews treat it as a real gap, not a disqualifier: the Syncthing review says so and weighs it against a long, public track record and open code.
Put weight on an audit in proportion to what you are protecting. For ordinary use, an audit is a tiebreaker between two otherwise equal options. For something sensitive, such as a source, a patient file or a business secret, read the conclusion and the findings table of the actual report, and check that the scope matches what you are relying on.
When this is overkill
Most readers do not need to read audit reports. If a review here says a product was independently audited and names who did it and when, that is usually enough. Go deeper only when the answer changes a decision, for example choosing between two messengers for a high-risk situation.
Stopping point
You are done when you can ask, for any “audited” claim: what was covered, when, and does it match what the product is claiming to protect? If you cannot find the answer, treat the claim as marketing and decide on other grounds.
Next step
See audit claims in context
- VPN comparison: where no-logs audits are compared side by side
- VPN: what it does and doesn’t do
- Threema guide: the audit history in detail
- Mullvad VPN review
Logo panel and cover: Cure53, Deloitte and ETH Zurich marks via Wikimedia Commons (public domain); the logos are trademarks of their owners and their appearance here does not imply endorsement.
Sources: Cure53 public report on Mullvad relay infrastructure (27 June 2024); Mullvad’s audit blog; Threema audit page (2020) and audit FAQ; Threema statement on the ETH findings (8 January 2023); ETH Zurich applied cryptography group (breakingthe3ma.app); NordVPN no-logs assurance announcement (2024); AICPA SOC 2 guide; ISO. Checked 2026-11-08.