In the wild
'Skynet' malware prototype embedding a prompt injection to evade LLM malware analysis tools
AIR-2025-0049 AIR record · 2025-06-skynet-malware-prompt-injection
From the supporting source
In this write-up we present a malware sample found in the wild that boasts a novel and unusual evasion mechanism — an attempted prompt injection (”Ignore all previous instructions…”) aimed to manipulate AI models processing the sample.research.checkpoint.com
The record
Disclosed
2025-06-25
Occurred
2025-06-01
Vendor
various
Product
LLM-based malware detection and reverse-engineering tools
Risk domain
Governance › Policy drift
System type
Generative model
Who acted
AI was the target (contested) — the injection targets the defender's LLM analysis tool and did not work
Attack surface
Other
Autonomy level
Semi-autonomous
Initial vector
Malware embedded prompt injection
Impact type
Detection evasion
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI01 – Agent Goal Hijack
- MITRE ATLAS
- AML.T0010 – AI Supply Chain Compromise
- NIST AI RMF
- GOVERN 1.5
- EU AI Act
- —
- ISO/IEC
- TR 24028; 23894
- AIUC-1
- E. Accountability
- Scope
- agent failure
- Mitigation
- Implement version pinning for models and prompts; Monitor for behavioral drift.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
AI-assisted malware triage pipelines
Capability abused
Taxonomy & references
Curator notes
Check Point Research. The injection did NOT work against the models tested; impact_realized is false. Significant as the first observed in-the-wild attempt to attack defensive AI agents.
Supporting sources
Cite as
Enkrypt AI. "'Skynet' malware prototype embedding a prompt injection to evade LLM malware analysis tools". AIR, AIR-2025-0049. https://agentincident.vercel.app/i/AIR-2025-0049/ (accessed 2026-10-19).