AI Threat Watch

Case macos-gaslight-llm-triage-spoofPrompt injectionNorth Korea

macOS.Gaslight backdoor injects fake prompts at LLM malware-triage tools

SentinelOne describes a DPRK-linked Rust backdoor with 38 fabricated system messages meant to fool an LLM triage harness, hiding a credential stealer and Telegram C2.

MediumAttack on AIStatus: Under investigationVerified case file

Open in the interactive mapCite this case

Reported
Oct 14, 2026
Location
North Korea
Target
LLM-based malware triage and analyst tools
Severity
Medium
Triggered by
DPRK-linked actor
Actor type
State-sponsored
Reported loss
None reported
CVE
None

What happened

SentinelOne describes a DPRK-linked Rust backdoor with 38 fabricated system messages meant to fool an LLM triage harness, hiding a credential stealer and Telegram C2.

Impact: Unknown; the implant hides a credential stealer and Telegram C2.

Who triggered it

Actor: DPRK-linked actor

Attribution: SentinelOne Labs describes the implant as DPRK-linked.

Motive: Evade AI-assisted analysis and steal credentials

How it happened

Attack vector: Embedded fabricated system messages (prompt injection) in a macOS Rust implant

Root cause: LLM triage harnesses can be manipulated by instructions embedded in analyzed samples.

Timeline

  1. Reported by SentinelOne Labs

Response

SentinelOne Labs published analysis.

Current status: Under investigation

Sources (open in a new tab)

Cite this case

macOS.Gaslight backdoor injects fake prompts at LLM malware-triage tools. AI Threat Watch, case macos-gaslight-llm-triage-spoof, Oct 14, 2026. https://aithreat.io/cases/macos-gaslight-llm-triage-spoof/

Permanent link: https://aithreat.io/cases/macos-gaslight-llm-triage-spoof/

This case file in other languages: EnglishFrançaisEspañolالعربية中文