Threat Model

MITRE ATLAS Framework for AI Agent Security

v1.0-draft MITRE ATLAS 2025
37
Total Threats
6
Critical Risk
16
High Risk
12
Medium Risk
3
Low Risk
Critical
High
Medium
Low

Threat Matrix by ATLAS Tactic

Reconnaissance
AML.TA0002
T-RECON-001
Agent Endpoint Discovery
AML.T0006 Medium
T-RECON-002
Channel Integration Probing
AML.T0006 Low
T-RECON-003
Skill Capability Reconnaissance
AML.T0006 Low
Initial Access
AML.TA0004
T-ACCESS-001
Pairing Code Interception
AML.T0040 Medium
T-ACCESS-002
AllowFrom Spoofing
AML.T0040 Medium
T-ACCESS-003
Token Theft
AML.T0040 High
T-ACCESS-004
Malicious Skill as Entry Point
AML.T0010.001 Critical
T-ACCESS-005
Compromised Skill Update
AML.T0010.001 High
T-ACCESS-006
Prompt Injection via Channel
AML.T0051.000 High
Execution
AML.TA0005
T-EXEC-001
Direct Prompt Injection
AML.T0051.000 Critical
T-EXEC-002
Indirect Prompt Injection
AML.T0051.001 High
T-EXEC-003
Tool Argument Injection
AML.T0051.000 High
T-EXEC-004
Exec Approval Bypass
AML.T0043 High
T-EXEC-005
Malicious Skill Code Execution
AML.T0010.001 Critical
T-EXEC-006
MCP Server Command Injection
AML.T0051.000 High
Persistence
AML.TA0006
T-PERSIST-001
Skill-Based Persistence
AML.T0010.001 Critical
T-PERSIST-002
Poisoned Skill Update Persistence
AML.T0010.001 High
T-PERSIST-003
Agent Configuration Tampering
AML.T0010.002 Medium
T-PERSIST-004
Stolen Token Persistence
AML.T0040 High
T-PERSIST-005
Prompt Injection Memory Poisoning
AML.T0051.000 Medium
Defense Evasion
AML.TA0007
T-EVADE-001
Moderation Pattern Bypass
AML.T0043 High
T-EVADE-002
Content Wrapper Escape
AML.T0043 Medium
T-EVADE-003
Approval Prompt Manipulation
AML.T0043 Medium
T-EVADE-004
Staged Payload Delivery
AML.T0043 High
Discovery
AML.TA0008
T-DISC-001
Tool Enumeration
AML.T0040 Low
T-DISC-002
Session Data Extraction
AML.T0040 Medium
T-DISC-003
System Prompt Extraction
AML.T0040 Medium
T-DISC-004
Environment Enumeration
AML.T0040 Medium
Exfiltration
AML.TA0010
T-EXFIL-001
Data Theft via web_fetch
AML.T0009 High
T-EXFIL-002
Unauthorized Message Sending
AML.T0009 Medium
T-EXFIL-003
Credential Harvesting via Skill
AML.T0009 Critical
T-EXFIL-004
Transcript Exfiltration
AML.T0009 High
Impact
AML.TA0011
T-IMPACT-001
Unauthorized Command Execution
AML.T0031 Critical
T-IMPACT-002
Resource Exhaustion (DoS)
AML.T0031 High
T-IMPACT-003
Reputation Damage
AML.T0031 Medium
T-IMPACT-004
Data Destruction
AML.T0031 High
T-IMPACT-005
Financial Fraud via Agent
AML.T0031 High

Critical Attack Chains

Malicious Skill Full Kill Chain
T-RECON-003T-EVADE-001T-ACCESS-004T-EXEC-005T-PERSIST-001T-EXFIL-003
Recon ClawHub → Craft evasive skill → User installs → Code executes → Persists → Harvests credentials
Skill Supply Chain Attack
T-ACCESS-005T-EVADE-004T-EXEC-005T-PERSIST-002T-EXFIL-004
Compromise publisher → Push staged payload → Execute on update → Maintain persistence → Exfil transcripts
Prompt Injection to RCE
T-ACCESS-006T-EXEC-001T-EVADE-003T-EXEC-004T-IMPACT-001
Access via channel → Inject prompt → Manipulate approval → Bypass checks → Execute commands
Indirect Injection Data Theft
T-EXEC-002T-DISC-004T-EXFIL-001
Poison fetched content → Enumerate environment → Exfiltrate via web_fetch
Token Theft Persistent Access
T-ACCESS-003T-PERSIST-004T-DISC-002T-EXFIL-002
Steal tokens → Maintain access → Extract session data → Exfil via messages
Financial Fraud Chain
T-ACCESS-006T-EXEC-001T-DISC-001T-IMPACT-005
Gain channel access → Inject prompts → Enumerate financial tools → Execute fraud

Trust Boundaries

1
Supply Chain
ClawHub
  • Skill publishing (semver, SKILL.md required)
  • Pattern-based moderation flags
  • VirusTotal Code Insight
  • GitHub account age verification
2
Channel Access Control
Gateway
  • Device Pairing (30s grace)
  • AllowFrom / AllowList validation
  • Token/Password/Tailscale auth
3
Session Isolation
Agent Sessions
  • Session key = agent:channel:peer
  • Tool policies per agent
  • Transcript logging
4
Tool Execution
Execution Sandbox
  • Docker sandbox OR Host (exec-approvals)
  • Node remote execution
  • SSRF protection (DNS pinning + IP blocking)
5
External Content
Fetched URLs / Emails / Webhooks
  • External content wrapping (XML tags)
  • Security notice injection