CVE-2026-54123

The CVSS 5.5 on this Defender for Endpoint information disclosure is misleading—not because the score is wrong for what it measures, but because it measures the wrong thing. An authorized local user querying Defender's sensor APIs can extract process behavior telemetry, credential access patterns, network connection data, and whatever organizational fingerprinting Defender has ingested about this endpoint. That's not a local file disclosure. That's the security team's own threat model, operationalized in telemetry, handed to whoever can query the sensor on an authorized context. The practical risk isn't the data on one machine. It's the strategic map of what the defender observes on that machine—which tells an attacker how to construct evasion for every machine Defender monitors. An insider or a compromised service account with local access now has the reconnaissance payload to understand the blue team's observational model and invert it. The 'authorized attacker' qualifier describes the default state of a persistent adversary; APT groups don't always need zero-days, they need to operate within products already deployed. What to do: audit which local users and service accounts can query Defender's sensor APIs, and determine whether that query surface was designed with the principle that authorized callers are untrusted relative to the sensor's own data. The patch addresses this specific flaw, but the broader pattern—security products whose privileged observation point becomes an information liability—has a lineage in Norton, McAfee, and HIPS products from the mid-2000s. This is that genetic sequence expressing itself again. Track these not just as individual CVEs but as a class where the blast radius extends laterally through the defender's own visibility.

Reviewed through automated stages and approved by a human before publication.