AEGIS · daily threat bulletin

What attackers are exploiting, distributing and controlling right now

Rebuilt every morning by a data pipeline that collects four public threat feeds, lands them in an Apache Iceberg lakehouse, models them with dbt and scores them with machine learning. There is no server behind this page: every number is a file the pipeline wrote.

Loading the latest bulletin…

CISA Known Exploited Vulnerabilities

Whose software is being exploited

Machine learning · text classifier

Recently exploited, and resembling ransomware targets

abuse.ch URLhaus

What malware is being handed out

Machine learning · clustering

Servers that behave like one operation

abuse.ch Feodo Tracker × Tor Project

Who is controlling the botnets

First-hand · Cowrie honeypot

Attacks on our own sensor

How this bulletin is made

From public feed to this page, every morning

  1. Collect. Four feeds are downloaded and published to Kafka (Redpanda), each record checked against an Avro schema.
  2. Land. A writer appends them to Apache Iceberg tables in S3-compatible storage, the Bronze layer: raw and never edited.
  3. Model. dbt builds deduplicated Silver tables and question-shaped Gold tables in DuckDB, with data tests that stop the build on a broken rule.
  4. Learn. scikit-learn clusters malware servers into campaigns and ranks new vulnerabilities, each model evaluated the way it is used.
  5. Publish. An allow-listed export writes these files. Honeypot addresses leave only as keyed pseudonyms.