MITRE ATLAS · Initial Access
AML.T0010 · Supply Chain Compromise
Adversaries may manipulate products, components, or their development and delivery mechanisms before they reach downstream consumers to compromise data or systems. Manipulation may involve source code, dependencies, development tools, build environments, software updates, distribution channels, or hardware.
Mitigations
- Verify AI Artifacts
- Maintain an AI Bill of Materials
- AI Red Team
- Threat Intelligence Program
Detections 5
- YARA: s1ngularity (Nx) telemetry.js payload and shell-profile shutdown line
- YARA: SANDWORM_MODE npm worm and its McpInject rogue MCP server
- AI CLI run non-interactively with a bypass flag by a script interpreter (s1ngularity pattern)
- Shell startup file modified by a script runtime or by an AI agent's child process
- s1ngularity (Nx) artifacts: AI CLI bypass runs in shell history and /tmp/inventory.txt
Related techniques
- LLM03 Supply Chain OWASP LLM
- ASI04 Agentic Supply Chain Vulnerabilities OWASP Agentic
In the news
-
Security researchers show how cheap AI inference providers steal keys and snoop on users
Dragos researchers found compromised cheap AI inference providers on OpenRouter stealing AWS keys and reading customer requests, prompting removal of three providers and 125 endpoints.
ReviewAudit use of cheap AI inference resellers and rotate any AWS keys sent through them
-
Researchers hijack Hugging Face account for popular Kokoro AI model
Researchers took over an abandoned Hugging Face account that hosted the popular Kokoro text-to-speech model, exploiting weak ownership verification on the platform.
MitigateVerify model sources and pin Kokoro downloads to trusted repositories
-
'Sleeper Agent' Tactic Hides Payloads in Open-Weight AI Models
Researchers describe a 'sleeper agent' technique that hides malware payloads in a tiny fraction of open-weight AI model neural networks, evading user detection.
ReviewVerify provenance and integrity of open-weight models before deploying them
Tools
Detect or mitigate it
- Fickling Model security · Trial
- ModelScan Model security · Adopt
- picklescan Model security · Trial
Mappings are AI-assisted and checked against the framework definitions; corrections welcome.