Ukraine solidarity. Service offered free to Ukrainian businesses for as long as the war lasts. Request free access
Security analysis - June 2026

Zero-Knowledge: what the June 2026 disclosures change (and do not change) for ARDNTECH

In June 2026, three events reignited the debate on the real guarantees of "zero-knowledge" managers: an academic study from ETH Zurich and Aalto University maps 25 attacks exploitable if the server is compromised, a breach at LastPass transits through a third-party CRM, and a DEF CON 33 researcher demonstrates the hijacking of autofill via clickjacking. Here is the concrete impact of these three events on the ARDNTECH architecture, without marketing rhetoric.

Published on 23 June 2026 by ARDNTECH EI  ·  Internal analysis based on the public disclosures

Context

Three disclosures in June 2026

12 June 2026 · Breach

LastPass via Klue - supply chain

A theft of OAuth tokens at Klue (a sales intelligence tool integrated with Salesforce) gave access to the contact data of LastPass customers. The encrypted vaults were not exposed. The real risk is targeted phishing based on a contact list stolen at a peripheral subprocessor, not the compromise of the cryptographic core.

Takeaway : zero-knowledge protects the vaults, not the metadata held by third parties.

June 2026 · Academic research

ZK study ETH Zurich + Aalto University

Matteo Scarlata and his co-authors (relayed by Ars Technica) map 25 attacks showing that the "zero-knowledge" guarantees of Bitwarden, Dashlane and LastPass degrade if the server is compromised and certain opt-in features are active: account recovery, sharing via public key distribution, cryptographic backward compatibility (KDF downgrade).

Takeaway : a formally correct ZK must withstand the scenario of a compromised server, not just an honest administrator.

June 2026 · DEF CON 33

Browser extension clickjacking

Marek Tóth demonstrates that the autofill of 11 managers can be triggered by an invisible overlay. A single click on a fake cookie banner is enough to fill the fields of a page controlled by the attacker. The surface affected is the DOM injection logic of the extension, not the server-side vault.

Takeaway : the browser extension is an attack surface distinct from the server, to be hardened independently.

Technical analysis

Mapping ETH/Aalto onto ARDNTECH

The study targets three families of features that, on a compromised server, weaken the ZK guarantees. Here is their correspondence in the ARDNTECH architecture.

Study vector ARDNTECH architecture Verdict
Account recovery
Copy of the vault key stored server-side, usable if the server is compromised
The wrap is produced client-side (Argon2id) from a recovery code that only the user holds. The server stores opaque wraps: it cannot re-derive the key. Consumed with immediate scrub. Sound
Cryptographic downgrade
Force a client to use a weaker algorithm during the recovery wrap
Blocked server-side: an inspection component parses the actual armor (SKESK v6, S2K Argon2id, AEAD present, floor costs). A tampered client that downgrades is rejected and forced into a reactivation. Neutralised
Public key substitution on sharing
The server serves its own key in place of the recipients
The server exposes the public key + fingerprint of the recipient. A June 2026 fix (extensions repo) covers client-side verification. Residual point: ensuring that a key change of a known contact triggers an explicit alert. Under verification
Admin recovery keys (LastPass type)
A group admin can re-encrypt a users vault server-side
ARDNTECH has no server-side "break-glass admin" mechanism that would re-encrypt a users vault key to an admin key. Team sharing goes solely through the teams public key and its fingerprint. Sound by design

Synthesis : a compromised ARDNTECH server does not allow decryption of the recovery wraps - they are opaque, derived client-side from a secret the server does not hold, and the KDF downgrade is refused at registration. These are precisely the scenarios exploited in other managers by the ETH/Aalto study. They are closed here by architectural choices documented in the threat model.

Browser extension

Clickjacking surface: the ARDNTECH extension

Tóths research targets the DOM injection logic of extensions. Autofill and autologin are the surfaces concerned. The server-side vault is not at fault - it is the extension filling the fields that can be tricked. The following points are being audited in the ARDNTECH extension.

  • Refusal on masked elements: no autofill if opacity below 1, visibility:hidden, near-zero size or covered element (hit-testing of the click point).
  • No autofill in a cross-origin iframe: verification that window.top === window.self or same origin.
  • <strong>Strict origin verification:</strong> exact match between the actual URL and the vault entry, with no lax substring matching.
  • <strong>Visible confirmation before injection:</strong> for autologin, no filling triggered by a click on an uncontrolled third-party element.
  • Injected interface not overlayable: isolated Shadow DOM, controlled z-index, visibility re-check at the moment of the click.
Supply chain

LastPass-Klue lesson: peripheral subprocessors

A sound cryptographic core can be bypassed by the peripheral SaaS stack. The LastPass vaults did not move, but a contact list was stolen at a third-party CRM integrator.

Minimise third-party SaaS

The current subprocessors of ARDNTECH (Scaleway, Brevo, Stancer, HIBP, Bugsink) are framed in the register published on the Trust page. Zero-knowledge protects the vaults, not a contact list in an unlisted third-party CRM.

OAuth token hygiene

Rotation of third-party integration tokens, minimal scopes, immediate revocation in case of incident. Access to peripheral systems must not allow a mapping of customers to be reconstructed.

Public subprocessor register

Every new subprocessor is subject to an update of the DPA before integration. The register is publicly accessible on the Trust page.

Private beta open

Test ARDNTECH with your team

Verifiable zero-knowledge, hosted in France, AGPLv3 source code. The private beta is accessible on sign-up.