-
Notifications
You must be signed in to change notification settings - Fork 0
Threat Model
This document describes the security assumptions, trust boundaries, potential attack vectors, and current limitations of MailIntel AI.
MailIntel AI is designed as an investigation and analysis platform. It assists investigators by collecting evidence and producing explainable reports. It is not intended to automatically classify or quarantine emails.
MailIntel AI aims to:
- Preserve investigation integrity.
- Produce reproducible results.
- Provide explainable findings.
- Prevent accidental execution of malicious content.
- Minimize trust assumptions.
The application separates trusted components from untrusted inputs.
User
│
▼
RFC822 Email (.eml)
(Untrusted Input)
│
▼
Email Parser
│
▼
Investigation Engine
│
▼
Evidence Collection
│
▼
Risk Engine
│
▼
Reports
(Trusted Output)
Only the uploaded email should be considered untrusted.
Important assets include:
- Uploaded email
- Extracted metadata
- Authentication headers
- Investigation evidence
- Generated reports
- Risk score
- Investigation history (future)
Maintaining the integrity of these assets is essential.
An attacker may craft malformed RFC822 messages intended to trigger parser failures.
Current mitigation:
- Standard Python email parser.
- Structured parsing workflow.
- Graceful exception handling.
Emails may contain:
- Phishing domains
- Typosquatted domains
- Tracking links
- Credential harvesting URLs
Current mitigation:
- URLs are extracted as text.
- Links are not automatically opened or executed.
Future enhancement:
- Threat intelligence reputation lookup.
Attackers frequently rely on persuasive language.
Examples:
- Urgent payment requests
- Account suspension warnings
- Password reset messages
- Executive impersonation
Current mitigation:
- Rule-based language analysis.
- Evidence-driven findings.
Attackers may spoof sender identities.
MailIntel AI evaluates:
- SPF
- DKIM
- DMARC
Authentication failures contribute to investigation risk.
Current version:
Attachment analysis is not implemented.
Uploaded email attachments are not executed.
Future versions may include:
- Attachment extraction
- File hashing
- Malware scanning
- Sandbox integration
MailIntel AI currently does not provide:
- Malware detection
- Antivirus scanning
- Endpoint protection
- Email quarantine
- Secure email gateway functionality
- Network intrusion detection
These capabilities are outside the scope of the project.
The system assumes:
- Uploaded emails are RFC822 compliant.
- Users are authorized to investigate submitted emails.
- Generated reports remain confidential.
- Host operating system is trusted.
Future versions may improve resilience against:
- Very large email files
- Archive extraction attacks
- Attachment-based malware
- HTML rendering exploits
- Embedded script analysis
- Image-based phishing
- QR code phishing
- Homograph domain attacks
MailIntel AI follows several core principles.
The application requests only the permissions required to perform analysis.
Every finding must be supported by collected evidence.
No hidden AI decisions are presented to investigators.
The same email should always produce the same investigation result.
This simplifies validation and auditing.
Email content is treated as untrusted input.
The parser extracts information without executing scripts, opening links, or rendering active content.
Potential future improvements include:
- VirusTotal integration
- AbuseIPDB integration
- Domain reputation services
- WHOIS enrichment
- DNS analysis
- Attachment malware scanning
- HTML email sanitization
- Sandbox execution
- Case management
- User authentication
- Audit logging
MailIntel AI is intended for:
- Security investigations
- Research
- Education
- Incident response
- Email analysis
Users remain responsible for validating findings before taking operational or legal action.
MailIntel AI provides investigative assistance and should not be considered a replacement for professional judgment.