Getting Started
What AIDE is
AIDE is the Global Cyber Alliance's honeypot network and the platform that reports on it. ProxyPot sensors present themselves as ordinary internet-facing services, record what arrives, and hand the result to AIDE. Every figure in the platform comes from a session a sensor actually recorded: the credentials that were tried, the commands that ran, the files that were fetched. Nothing is modelled, sampled or inferred.
The corpus runs to billions of sessions and starts in September 2018.
What a sensor records
Sensors present seventeen protocols, in both IPv4 and IPv6:
| Protocol | What arrives |
|---|---|
| Telnet, Telnets | IoT and embedded targeting, overwhelmingly with default credentials |
| SSH, SFTP | Shell access attempts, credential stuffing, tunnel requests |
| FTP, FTPS | File transfer abuse and brute force |
| HTTP, HTTPS | Web exploitation and vulnerability probing |
| SMTP, SMTPS | Spam and open-relay probing |
| SOCKS5, HTTP CONNECT | Open-proxy abuse: requests to relay traffic onward |
| DNS, DoH, DoT | Open-resolver abuse and DNS exfiltration |
| TCP | Raw connections to a port with no application protocol behind it |
| ICMP | Reachability sweeps |
Not every sensor runs every protocol: see sensor profiles.
Telnet and SSH dominate, which is itself the insight: the internet is still being attacked through remote-access protocols designed in the 1980s.
A sensor never establishes an outbound connection on an attacker's behalf. Tunnel requests, proxy requests and DNS exfiltration attempts are recorded and refused, so the data represents intent rather than a relay that worked.
What a session is
One session is one interaction, and it is the unit everything else is built from. A session carries its own timing, the sensor that recorded it, the attacker's address and its geo and network attribution, and whatever the attacker did: credentials, commands, URLs, file hashes, HTTP requests, DNS queries and tunnel requests.
Every session is classified into exactly one category, in escalation order:
| Category | Shown as | Meaning |
|---|---|---|
scan | Scan | Connected and left. No credentials offered. |
attempt | Scouting | Credentials were tried and none worked. |
intrusion-login | Credentials worked | Credentials worked. |
intrusion-commands | Code pushed | Credentials worked and commands were run. |
intrusion-control | Command and control | The host was taken under external control. |
The API and exports carry the raw value; the web interface shows the label.
The three intrusion-* categories together are the number worth quoting: over half of all sessions in
the corpus reach one of them.
Because each session lands in exactly one category, the shares always sum to 100%, and the funnel on the dashboard is a partition rather than a set of overlapping counts.
Getting access
Access is granted per organization, with a data-access level agreed at the time. Request it from the landing page, or contact GCA directly.
What an account adds over the public landing page:
- IP-level data. Which addresses attacked, and when.
- Network attribution. Which autonomous system the traffic came from, and its operator.
- Content. The usernames, passwords, commands, URLs and file hashes actually seen.
- Discover. Query the session table directly by any allowed field.
- Exports. CSV of any query, up to five million rows.
- AS reports. A monthly document on any autonomous system.
The public landing page is deliberately limited to aggregate, economy-level figures. It carries no address, no network attribution and no content, and that is enforced by serving it only from the daily rollups, none of which hold those columns.
Using the API
Everything the web interface does is an API call. Sign in, create a key on the API Keys page, and send it as a bearer token:
curl -H "Authorization: Bearer $AIDE_KEY" https://gcaaide.org/api/metrics
A key carries its owner's group and their organization's data access, so it can do no more than they can. See Authentication for the details, and the API tab above for the full reference.