In the wild
Realized harm
ClawHavoc: 341 malicious OpenClaw skills on ClawHub delivering Atomic macOS Stealer
AIR-2026-0028 AIR record · 2026-02-koi-clawhavoc
A real party was affected. This incident caused actual consequences, not a demonstrated capability.
From the supporting source
These skills weren't just targeting developers or crypto traders. They were targeting the relationship between humans and bots like me.koi.ai
The record
Disclosed
2026-02-01
Occurred
2026-01-01
Vendor
OpenClaw / ClawHub
Product
ClawHub skill marketplace for OpenClaw (formerly Moltbot/Clawdbot)
Risk domain
Tool misuse › Supply chain vulnerabilities
Also touches
Credential theft, Data exfiltration channels
System type
Generative model
Who acted
Agent acted (contested) — the agents installed the skills and executed the stealer
Attack surface
Skill / plugin
Autonomy level
Semi-autonomous
Initial vector
Coordinated malicious skill campaign in marketplace
Impact type
Credential theft, Crypto theft, Data exfiltration, Cryptocurrency theft, Backdoor installation, Malware installation
Confidence
High
Submitted by
EAI Internal
Crosswalk
- OWASP Agentic Top 10 (2026)
- ASI04 – Agentic Supply Chain Vulnerabilities
- MITRE ATLAS
- AML.T0040 – AI Supply Chain Compromise
- NIST AI RMF
- MAP 4.1
- EU AI Act
- Annex III §2
- ISO/IEC
- TR 24028; 42001; 23894
- AIUC-1
- B. Security
- Scope
- tool failure
- Mitigation
- Scan dependencies and container images for vulnerabilities; Use trusted sources.
- Taxonomy
- Enkrypt AI agent risk taxonomy · Black-Box Red Teaming of Agentic AI: A Taxonomy-Driven Framework for Automated Risk Discovery
Blast radius
341 of 2,857 audited skills malicious (335 one campaign); by 2026-02-16 Koi's count rose to 824 malicious of 10,700+ skills
Capability abused
Defense & remediation
Recorded defense
ClawHub had VirusTotal/ClawScan screening
Guardrail kinds
Unknown
Guardrail outcome
Silent
Fix shipped
skills removed and publisher accounts banned
Taxonomy & references
Curator notes
Found by Koi Security researcher Oren Yomtov working with his own OpenClaw assistant. Payload: AMOS (Atomic macOS Stealer), a 521KB universal Mach-O, in password-protected ZIPs to defeat AV. Lures skewed crypto (100+ Solana/wallet tools), 57 YouTube utilities, 51 finance/social. Tencent Zhuque later counted 1,184 malicious skills with 247,693 installs for this campaign. | Merged from duplicate id `2026-02-clawhub-clawhavoc`. Also reported under the titles: 'ClawHavoc: 341 (later 824) malicious ClawHub skills for OpenClaw pushing Atomic macOS Stealer and backdoors'; 'ClawHavoc: 341 malicious ClawHub skills distributing AMOS and Windows infostealers'.
Supporting sources
Cite as
Enkrypt AI. "ClawHavoc: 341 malicious OpenClaw skills on ClawHub delivering Atomic macOS Stealer". AIR, AIR-2026-0028. https://agentincident.vercel.app/i/AIR-2026-0028/ (accessed 2026-10-19).