Skip to main content

Methodology

How we collect, enrich, and deliver threat intelligence

Source Selection

We aggregate from 73+ sources across 11 categories: news, vendor research, EU government advisories, law enforcement, OSINT, CVE feeds, and more. Every source is public, auditable, and RSS-based.

AI Enrichment Pipeline

Each article passes through a multi-stage AI pipeline:

Stage 1: Basic Classification

Workers AI, Llama 3.3 70B — threat level, summary, key points

Stage 2: Deep Enrichment

OpenAI GPT-4o — MITRE ATT&CK mapping, NIS2/DORA compliance tagging, threat actor attribution, IOC extraction, sector classification

Stage 3: Multi-Source Synthesis

Related articles are clustered and synthesised into intelligence reports

MITRE ATT&CK Mapping

Techniques are identified using keyword matching validated by GPT-4o analysis. We map to both parent techniques (e.g., T1566) and sub-techniques (e.g., T1566.001). Mapping confidence varies — always verify against primary sources.

Compliance Tagging

NIS2 and DORA tags are applied at the article level by GPT-4o based on the content's relevance to specific regulatory articles. These are AI-generated suggestions, not legal determinations.

Threat Actor Attribution

Actor names are matched against a database of 45+ tracked groups using direct mentions, known aliases, country-level attribution, and TTP pattern matching. Confidence levels (high/medium/low) indicate the reliability of each attribution.

Data Freshness

Sources are collected hourly. AI processing runs immediately after collection. The entire pipeline from RSS ingestion to enriched intelligence takes approximately 15 minutes.

Disclaimer

All AI-generated analysis (threat levels, MITRE mappings, compliance tags, actor attributions, IOC extraction) should be independently verified before use in compliance evidence, incident response, or regulatory submissions.