CVE-2026-77332

memtomem-stm memories could escape their rendered data boundary

memtomem · memtomem-stm

First documented disclosure · Review checkpoint

Summary

Untrusted memory fields could alter the structure of context presented to an agent.

Severity and exploitation evidence

Low · 2.3/10 · CVSS 4.0. Severity source.

Exploitation not established: Public evidence describes a weakness, source correction or controlled demonstration; malicious exploitation against users is not established.

Exploitation evidence source

Technical severity is not a measure of observed attacks or of risk to your deployment. Attempts, demonstrations, and successful compromise retain the qualifications in the source.

How the weakness works

Markdown/HTML-compatible memory values were inserted without inert serialization.

Conditions required

Attacker control of long-term or working-memory data consumed by the agent.

Potential impact

Misleading context boundaries or concealed content can influence agent interpretation.

Affected versions and configurations

0.1.0–0.1.33.

Fix and mitigation status

0.1.34.

Practical next steps

Upgrade; treat all retrieved memories as untrusted data.

Operating lesson

Serialize retrieved fields as data and validate copied identifiers.

Evidence and limits

The fix addresses structural serialization; ordinary natural-language persuasion remains a residual model risk. CNA publication 2026-11-20 is distinct from the earliest reviewed public source. GitHub_M assessment: CVSS 4.0 2.3.

Sources