Edgeless Systems Contrast EUVD-2026-87832

Requires a specific misconfiguration plus attacker positioning: (1) the operator must configure a per-registry block in the imagepuller using an FQDN with NO leading dot, e.g. [registries."ghcr.io."] or [registries."ghcr.io"] - configurations using a leading dot such as [registries.".example.registry"] are entirely unaffected; and (2) an image or layer must be pulled from a sibling domain whose hostname ends in that byte sequence (e.g. evilghcr.io), which the attacker must register/control, or which occurs coincidentally. Limiting factors: dual preconditions (operator config error AND attacker-controlled sibling domain in the pull path) make exploitation high-complexity; only Contrast versions through 1.20.0 are affected (fixed in 1.21.0, which enforces exact per-DNS-label equality); and the leak is bounded to credentials/CA-trust/mirror behavior, not image contents.

This is a genuinely low-to-moderate real-world risk best treated as configuration hygiene rather than an urgent priority. The vendor CVSS 4.0 vector (6.3, AV:N/AC:H/AT:P/PR:N/VC:L) is internally consistent with the described mechanics: network vector because credentials leak to an attacker-controlled registry, but AC:H and AT:P because exploitation demands two independent preconditions - the operator must have configured a registry FQDN WITHOUT a leading dot, AND the attacker must control a sibling-suffix domain (e.g. evilghcr.io) that the deployment actually pulls an image or layer from. Impact is confined to confidentiality (VC:L / C:L): leaked Authorization header plus disabled TLS verification and mirror redirection for the sibling host. There is no integrity or availability impact - image bytes stay pinned by digest and are validated after pull, so no code substitution is possible. No EPSS score, no CISA KEV listing, and no public POC are present in the input; the advisory describes the flaw clearly but ships no weaponized exploit. Signals do not conflict: low likelihood (high complexity, dual precondition, no exploitation evidence) matches the moderate-but-not-critical score. The only notable metadata gap is that this is credential disclosure to a system under attacker control, which some analysts would model as a scope change (S:C) - the vendor chose S:U, keeping impact within the imagepuller.

An operator configures a Contrast imagepuller registry block as [registries."ghcr.io."] (no leading dot) with a registry token and custom CA. An attacker registers evilghcr.io and gets the deployment to reference an image or layer on that host; because HasSuffix matches without a label boundary, the imagepuller sends the configured Authorization header (and honours insecure-skip-verify) to the attacker's server, leaking the credential. No public exploit code is identified at time of analysis.