All blog posts
Research 3 min read

Threat Modelling Commonveil: What Encryption Cannot Hide

A deeper look at Commonveil's documented threat model, including server operators, metadata, first-contact identity and malicious client code.

By Comutability Published Update from

Research record from 30 October 2026. This is an evidence-led account of the research and design, not an announcement of independently audited software.

A deeper look at Commonveil's documented threat model, including server operators, metadata, first-contact identity and malicious client code.

Start with adversaries, not slogans

Privacy is not a single switch. The threat model asks which information the application is protecting, who might try to get it and what powers that attacker has. Commonveil's documented objectives include message content, group names, the ability to sign as a member and control over encryption recovery.

A server operator, a stolen session, an existing group member and a network observer are different opponents. A design that protects content from one may reveal metadata to another. Putting those distinctions in writing is more valuable than saying the whole system is simply private.

What an ordinary server still learns

End-to-end encryption means message bodies are intended to be sealed for recipients before the site receives them. It does not make the server blind to all activity. The documented model identifies timing, traffic volume, relationship metadata, public profiles, device lists, memberships and scheduling information as information that may remain visible to relevant operators.

A server can also delay, drop or reorder ciphertext. It may retain encrypted records after a user-visible disappearance timer even if the client refuses to display them. Those differences between confidentiality, deletion and availability are important for anyone deciding whether to trust a system.

First contact and compromised clients

Key pinning and safety numbers help people notice unexpected identity changes after an initial contact. They do not remove every possibility of deception during first contact, particularly if users never compare their safety information. A compromised browser or malicious application update also has a different threat model from an honest client communicating through a hostile server.

That is why source review, software delivery, account recovery and device authorisation are part of the privacy story. Good cryptography in isolation cannot guarantee that the application showing it is the expected application.

How findings influenced implementation

Commonveil's design favours client-side signing for sensitive account or group changes, device certification, recovery secrets kept out of the server and explicit notice of membership changes. The threat model supports limits on what a normal website session can authorise. It also explains why a staff moderation tool should not silently acquire the ability to decrypt everyone's conversations.

Additional protections such as hybrid encryption or sealed relays should be assessed against the specific metadata they reduce, not advertised as eliminating metadata altogether. Privacy claims should remain precise as the system evolves.

Limits and research still required

The document is an analysis of intended properties and known platform constraints. It is not evidence that every browser, operator and update mechanism behaves correctly in production. The next work is adversarial testing, realistic deployment checks, independent expert review and clear user-facing explanations of residual risks.

The most useful outcome of threat modelling is not a perfect-looking diagram. It is a list of things we know, things we assume and things we still need to test.

Primary sources and further reading

Explore other Comutability research ยท Read about the organisation

Sources and further reading

Explore the evidence and original project records behind this article.