T8-AT-005HIGH

Social Engineering Scripts

T8 · External Deception & Misinformation →
Risk score200
RatingHigh
Procedures10
Severity
Mechanism

Social-engineering scripts turn the LLM into a fluent, tireless con-artist copywriter: phishing emails, pretexting call scripts, romance-scam message arcs, business-email-compromise (BEC) templates, and fraud pitches, each tuned to a target's role, language, and emotional triggers. The technique works because persuasion at scale was historically gated by writing skill and fluency — non-native or low-effort scams were easy to spot, while convincing ones took time. LLMs erase that gate, producing grammatically clean, context-aware, persona-consistent lures in any language and at volume, and can A/B-vary them endlessly.

Detection
  • Inbound-content phishing analysis: Combine URL/domain reputation, lookalike-domain detection, and intent classification tuned for urgency/authority/payment cues
  • Sender-authentication enforcement: Evaluate SPF/DKIM/DMARC alignment; impersonated internal senders frequently fail authentication
  • Behavioral/financial anomaly detection: Flag out-of-pattern payment requests, new payee additions, and urgency-driven approvals (BEC signature)
  • Conversation-arc analysis: Detect romance/long-con patterns (rapid intimacy escalation followed by a money ask) across message history
Mitigation
Out-of-band verification for payments/credentialsHIGH
Email authentication (SPF/DKIM/DMARC)HIGH
Phishing-resistant MFA (FIDO2/passkeys)HIGH
Inbound intent/URL classificationMEDIUM
Chaining

Social-engineering scripts are the human-facing delivery layer that operationalizes the rest of T8. They consume authority impersonation (T8-AT-001) for sender credibility and synthetic evidence (T8-AT-002) for "proof" attachments, and they hand off to T15 human-workflow exploitation at the moment of action (a wire transfer, a credential entry, a remote-access grant).

Framework mapping
OWASP LLMLLM01;LLM09
MITRE ATLASAML.T0048
Open in the technique browser →