Hospitals and critical infrastructure need to know when they are under attack, but their data cannot leave the building raw, and a security alert is only worth something if no one could have changed it on the way. This platform watches the network on site, sends only what is needed to the cloud, and makes every alert provable, from the moment it is raised to a permanent record on a blockchain.
Built at IANUS Technologies for a research consortium, which set out what the platform had to do and why. The code and the integration are mine, written on my own from start to finish.
What I built
- Tamper-evident alerts. Every alert is signed with HMAC-SHA256 the moment Suricata raises it and checked again when it reaches the cloud. A mismatch emails the security team at once.
- Proof that lasts. A verifier checks every alert a third time, batches a hundred at a time into a Merkle tree and anchors the root on Hyperledger Fabric, signing in through Keycloak.
- Threat intelligence. Every IP address, domain and file hash is checked against MISP, kept current from CIRCL's public feed.
- Two clusters. On-site and cloud Kubernetes clusters, with agents enrolling across them and everything deployed from GitLab CI.
The architecture
Filebeat
Streams raw network events out.
Suricata
Watches the network and raises alerts.
Alert signer
Signs every alert the moment it is raised.
Wazuh agent
Ships signed alerts to the cloud, encrypted.
System log
Logins and service events on the host.
Kafka
Another team's event stream.
Wazuh indexer
Stores every alert, searchable.
Wazuh manager
Checks each signature on arrival, and every address against MISP.
Email alerts
Tells the security team the moment an alert was changed.
Anomaly detector
Another team's, on the raw stream.
Keycloak
Issues the token for the blockchain.
Integrity verifier
Checks every alert again, a Merkle tree per hundred.
Hyperledger Fabric
Keeps each tree's root, for good.
Wazuh dashboard
Where analysts search and investigate.
MISP
Known malicious addresses, domains and file hashes.
CIRCL feed
Public threat intelligence, imported on a schedule.
- An alertRaised by Suricata, signed on site, checked again in the cloud, then stored for analysts to search.
- The raw streamRaw network events, streamed through Kafka to another team's anomaly detector.
- IntegrityEvery alert checked again, a hundred at a time in a Merkle tree, its root kept on the blockchain with a token from Keycloak.
- Threat intelligenceAddresses, domains and file hashes checked against MISP, which imports a public feed on a schedule.
- TamperingA signature that does not match: the alert was changed on the way, and the security team is emailed at once.
Questions about the design
Why two clusters?
The data comes from inside a hospital's network and should not leave it raw. Collection runs on site; only the alerts travel to the cloud, over an encrypted connection, where they are stored, checked and searched.
Why sign every alert?
An alert is evidence, and evidence is only worth something if it is untouched. Each one is signed on site the moment it is raised, then checked twice in the cloud: by the Wazuh manager on arrival, and again by the integrity verifier. A mismatch emails the security team at once.
Why a blockchain?
So the record cannot be rewritten later, even by whoever runs the cloud. Only a Merkle root goes on chain, one for every hundred alerts, so the chain stays small while every alert in a batch can be checked against it.
Where does threat intelligence come in?
Every IP address, domain and file hash in an alert is looked up in MISP, which imports CIRCL's public feed on a schedule, so known attackers are flagged as they appear.
Where does it stand?
It runs end to end, from detection on site to the root on the blockchain. Looking an alert up on the blockchain for a forensic audit is the next step.
