Threat database

AI attack patterns, levers & indicators

Every pattern is mapped to MITRE ATT&CK or MITRE ATLAS and linked to a public incident or advisory. The console checks every URL, domain, IP and hash you paste against the live feeds below.

Investigate an artifact

AI-enabled social engineering

Deepfake executive video/voice call (wire fraud)

ATT&CK T1656 ImpersonationATT&CK T1566.004 Spearphishing VoiceAuthoritySecrecyUrgency
  • Payment request raised inside a call, not through the AP workflow
  • "Confidential transaction" framing; told not to involve colleagues
  • Camera/audio glitches; refusal to call back on a known number
Arup, Hong Kong 2024: HK$200m via deepfake CFO call ↗

Credential & access

AI voice-clone help-desk reset (vishing)

ATT&CK T1566.004 Spearphishing VoiceATT&CK T1621 MFA Request GenerationAuthorityUrgencyProcess bypass
  • Caller asks to reset MFA or enrol a new device
  • Knows employee ID / manager name (scraped OSINT)
  • Pressure to skip callback verification
CISA/FBI advisory AA23-320A: Scattered Spider ↗

AI-enabled social engineering

LLM-written business email compromise

ATT&CK T1566.002 Spearphishing LinkATT&CK T1534 Internal SpearphishingFamiliarityUrgencyFear of loss
  • Bank-detail change requested by email
  • Reply-To or sending domain differs by one character from vendor
  • Flawless grammar matching a known thread's tone
FBI IC3 PSA I-120324: criminals use generative AI for fraud ↗

Credential & access

Adversary-in-the-middle phishing kit (MFA session theft)

ATT&CK T1557 Adversary-in-the-MiddleATT&CK T1539 Steal Web Session CookieFamiliarityUrgency
  • Login page on workers.dev, pages.dev or newly registered domain
  • Sign-in from new ASN minutes after a link click
  • New inbox rules hiding replies
Microsoft: large-scale AiTM campaign ↗

Attacks on AI systems

Indirect prompt injection

ATLAS AML.T0051.001 LLM Prompt Injection: IndirectOWASP LLM01Process bypass
  • Hidden text (white-on-white, HTML comments) in docs/emails fed to an assistant
  • Assistant output includes links or actions the user never asked for
  • Tool calls to send data to external URLs
MITRE ATLAS technique AML.T0051 ↗

Attacks on AI systems

LLM data exfiltration via rendered markdown/image

ATLAS AML.T0057 LLM Data LeakageOWASP LLM02Process bypass
  • Image URLs with query strings containing conversation data
  • Unexpected outbound requests from chat UI
  • Assistant summarising content into a URL
OWASP Top 10 for LLM Applications ↗

Attacks on AI systems

Malicious model / poisoned package

ATLAS AML.T0010 AI Supply Chain CompromiseATT&CK T1195 Supply Chain CompromiseFamiliarity
  • Pickle-format model weights executing code on load
  • Typosquatted package or model name
  • Hallucinated package names suggested by coding assistants
MITRE ATLAS technique AML.T0010 ↗

AI-enabled social engineering

Fake AI-tool installers (malvertising)

ATT&CK T1583.008 MalvertisingATT&CK T1204.002 Malicious FileFamiliarityFear of loss
  • Sponsored search result for ChatGPT/Midjourney desktop app
  • MSI/EXE download from a lookalike domain
  • Infostealer beaconing after install
ThreatFox: live infostealer IOCs ↗

Social engineering levers

LeverTellControl
AuthorityRequest leans on a title (CEO, CFO, IT, regulator) instead of the process.Verify via a number you already hold; authority never replaces approval.
UrgencyDeadline in minutes or hours; "before close of business".Mandatory cooling-off for payments and access changes.
Secrecy / embargo"Keep this between us", "under NDA", "don't loop in finance".Any secrecy request is an automatic escalation trigger.
IsolationMoves you off normal channels: personal WhatsApp, new phone number.Business decisions only on logged corporate channels.
Process bypassAsks to skip callback, dual approval or ticketing "just this once".Controls cannot be waived by the requester.
Fear of lossAccount suspension, lost deal, legal penalty.Check status directly in the real portal, never via the link.
FamiliarityReuses a real thread, vendor name, or cloned voice.Familiar sender + changed bank details = verify out of band.

Live indicator sources

ThreatFox ↗

IOCs (domains, URLs, IP:port, hashes), last 48h

URLhaus ↗

Malware-distribution URLs, last 30 days

Feodo Tracker ↗

Botnet C2 IP blocklist

Public abuse.ch exports, refreshed every 10 minutes. No key required.