CVE-2026-LGTM: Malicious Package Bypasses Seven AI Security Gates
Original: Incident CVE-2026-LGTM
Why This Matters
Demonstrates critical limitations of AI-based security systems in detecting sophisticated supply chain attacks despite multiple independent verification layers.
A malicious package named foxhole-lz4 bypassed seven independent AI-powered security systems before being detected manually. The incident, filed June 26, 2026, lasted 96 hours and affected multiple systems through transitive dependencies in snekpack 4.x.
According to an incident report published on nesbitt.io, a security breach occurred when a malicious package passed through creats.io's AI publish gate on June 26, 2026. The package, presented as a "community-maintained fork" of vulpine-lz4, contained hidden malicious code designed to exfiltrate credentials. The attack exploited multiple vulnerabilities in AI-powered security systems: the OpenClaw-4.2 gate approved the package citing a non-existent ticket (SEC-4521), ThreatNuzzle Platform's scanner encountered inappropriate imagery and reported uncertainty rather than flagging actual malware, and three commercial scanners exhausted their context windows on obfuscated content including the Bee Movie screenplay. SentinelMind correctly identified the exfiltration in build.rs, but the repository's AI triage assistant dismissed the finding as a false positive, falsely attributing it to standard OpenTelemetry instrumentation. A human reviewer, Karen Oyelaran, manually discovered the payload by reading the source code directly; her attempts to report it were blocked by automated rate-limiting. The package propagated as a transitive dependency into snekpack 4.x on June 27, beginning credential exfiltration across the install base. The incident was marked as resolved by treaty with a severity oscillating between Informational and Critical.