MITRE ATLAS AI Threat Detection for Splunk

MITRE ATLAS is the adversarial threat matrix for AI/ML systems — the AI equivalent of MITRE ATT&CK. As organizations deploy LLMs, RAG pipelines, and ML APIs into production, they create an attack surface that most Splunk deployments have zero detection coverage for. This app provides 10 detection rules for AI and LLM threats — 9 mapped to specific MITRE ATLAS technique IDs, plus 1 vendor-defined rule where no canonical ATLAS technique covers the detection concept. Each rule is a Splunk saved search that monitors your AI/LLM logs for known attack patterns including prompt injection, jailbreak attempts, data exfiltration via inference APIs, training data poisoning, model reconnaissance, and AI abuse. The app includes a guided Setup dashboard that auto-discovers AI/LLM data in your environment, validates required fields, and shows which rules your data supports. A Configuration Guide provides platform-specific instructions for 12 LLM providers including LiteLLM, Azure OpenAI, AWS Bedrock, OpenAI, GCP Vertex AI, Kong AI Gateway, Portkey, Helicone, Cloudflare AI Gateway, Anthropic, self-hosted models (Ollama, vLLM, TGI), and custom API gateways. Rules are organized into two tiers: Tier 1 (Operational) — works with standard telemetry: token counts, API call volumes, storage access logs. Available from most platforms with default logging. Tier 2 (Content Inspection) — requires actual prompt/response text in log events. Requires explicit opt-in on all major platforms. The Configuration Guide explains how to enable this for each provider. Detection Coverage: AML.T0051.000 Direct Prompt Injection (Tier 2) AML.T0051.001 Indirect Prompt Injection via Retrieved Content (Tier 2) AML.T0054 LLM Jailbreak (Tier 2) AML.T0024 Exfiltration via AI Inference API (Tier 1) AML.T0020 Training Data Poisoning (Tier 1) GIC-VD-001 AI-Enabled Bulk Content Generation — vendor-defined, no canonical MITRE ATLAS technique covers this concept (Tier 1) AML.T0048 External Harms (Tier 1) AML.T0012 Valid Accounts (Tier 1) AML.T0014 Discover AI Model Family (Tier 2) AML.T0007 Discover AI Artifacts (Tier 1) All rules ship disabled by default. No Python scripts, no external dependencies — pure SPL and CSV lookups. Detection results are written to the summary index for fast dashboard rendering. Built by a Splunk Enterprise Architect with 10+ years of hands-on experience. This app is a standalone community release from GIC Engineering Consultants. Looking for broader AI/LLM threat coverage alongside SBOM security and CVE correlation? I also published a Supply Chain & AI Threat Intelligence Platform (SCIP) app: splunkbase.splunk.com/app/8814. Its free tier includes 10 comparable AI/agentic threat rules (9 MITRE ATLAS-mapped) at no cost. Paid tiers include 66 AI/agentic threat rules (57 MITRE ATLAS-mapped), plus additional capabilities.