The anti-abuse system detects abuse on the proxy fleet (exploit scanning, brute force, spam, credential stuffing) and publishes the result as a policy: a versioned list of rules plus quarantine triggers. The proxy (Transit) does not define rules itself. It pulls the policy, checks every request against it before dialing, and writes the matching rule id to its access log.
Base URL: https://aisec.orb-labs.com/v1. All calls except /healthz and /docs
need Authorization: Bearer <token>.
| Call | Returns |
|---|---|
GET /v1/policy | The current policy (JSON, below). Sends an ETag; send it back as
If-None-Match to get 304 Not Modified when nothing changed. |
GET /v1/policy/meta | Version, expiry, rule counts per kind, size, ETag. Cheap check. |
GET /v1/healthz | {"ok":true} |
GET /v1/docs | This page. |
Errors: 401 bad or missing token, 503 no policy available yet.
curl -s -H "Authorization: Bearer $TOKEN" https://aisec.orb-labs.com/v1/policy/meta
curl -s -H "Authorization: Bearer $TOKEN" -H 'If-None-Match: "<etag>"' https://aisec.orb-labs.com/v1/policy -o policy.json -w '%{http_code}\n'
If-None-Match), from the fleet
app, and push it to the Transit hosts. Hosts load it with hot reload, like blacklist_regex_file:
if a file is invalid, keep the previous one.expires (unix seconds) is about 2 hours after it was generated. If a host has no newer policy by then, it
drops all rules and triggers (fail open). That way a stale policy never keeps blocking.{
"v": 1,
"version": "20261119T101000Z",
"generated": 1763547000,
"expires": 1763554200,
"rules": [
{"id": "bl:malware-host.example", "kind": "blocklisted_host", "action": "log",
"hosts": ["malware-host.example"], "expires": 0},
{"id": "smtp:gw_203.0.113.7", "kind": "smtp_source", "action": "log",
"src": ["203.0.113.7"], "ports": [25, 465, 587], "expires": 0},
{"id": "ex:lv123:VPN_appliance", "kind": "exploit_source", "action": "log",
"src": ["198.51.100.0/24"], "ports": [10443, 4443, 8443], "expires": 0}
],
"quarantine": { ... section 5 ... }
}
| Field | Meaning |
|---|---|
id | Stable rule id. Characters A-Za-z0-9:._- only, no spaces, so it is safe in log lines. |
kind | blocklisted_host, smtp_target, exploit_source,
smtp_source, credential_stuffing_source. For stats only, no effect on matching. |
action | log or block (see below). |
src | Client IPs / CIDRs: the IP that connects to Transit (customer IP, or the gateway IP for gateway traffic). |
user | Proxy usernames (optional; not used yet). |
hosts | Exact target host names (lowercase, no trailing dot). |
suffixes | Domain suffixes: example.com matches example.com and *.example.com. |
dst | Target IPs / CIDRs (an IP-literal target, or the resolved IP). |
ports | Target ports. |
expires | Unix seconds; 0 = valid until the policy expires. |
Matching: a rule matches when every field it sets matches (AND). Inside one field, any value matches (OR). A field that is not set matches everything. Check every request (CONNECT, SOCKS, HTTP) before dialing, on gateway ports and direct (station) ports.
Action:
block: refuse the request (HTTP 403 / SOCKS rule-set failure) and log it with the rule id.log: let the request through unchanged and log it as if it had been blocked:
would_block rule=<id>. That way a new rule can run on live traffic, and we can check who it would
hit (false positives, volume) before it blocks anything. Switching a rule from log to
block is only a change in the policy, not in Transit.Today every rule is log. We switch rules to block once the log results are reviewed.
Access log: add two fields to the request line: rule=<id> (first matching rule, or
-) and enf=block|log|quarantine|-. Also log one line per policy load with the version and
the number of rules.
A quarantine denies all requests from one client (key src = client IP, or user)
for a time. Transit counts the metrics itself, because only Transit sees every request. The policy carries the
thresholds:
{
"action": "log",
"default_timeout_s": 3600,
"manual": [],
"triggers": [
{
"desc": "upstream connect failures (refused, timeout, unreachable, DNS error)",
"id": "q-errors",
"key": "src",
"metric": "failed_connects",
"threshold": 1000,
"window_s": 1800
},
{
"desc": "distinct destination IPs on attack ports (scanning)",
"id": "q-sweep",
"key": "src",
"metric": "distinct_dst_ips",
"ports": [
23,
102,
135,
139,
389,
445,
502,
636,
943,
1433,
1883,
2083,
2087,
2323,
2375,
2376,
2379,
3306,
3389,
3443,
4433,
4443,
5432,
5443,
5900,
5985,
5986,
6379,
7001,
7002,
7443,
8009,
8088,
8291,
8443,
8500,
8834,
9200,
9443,
10250,
10443,
11211,
27017,
50000,
60443
],
"threshold": 300,
"window_s": 600
},
{
"desc": "connections to one destination IP:port on attack ports (brute force)",
"id": "q-hammer",
"key": "src",
"metric": "conns_per_dst",
"ports": [
23,
102,
135,
139,
389,
445,
502,
636,
943,
1433,
1883,
2083,
2087,
2323,
2375,
2376,
2379,
3306,
3389,
3443,
4433,
4443,
5432,
5443,
5900,
5985,
5986,
6379,
7001,
7002,
7443,
8009,
8088,
8291,
8443,
8500,
8834,
9200,
9443,
10250,
10443,
11211,
27017,
50000,
60443
],
"threshold": 300,
"window_s": 600
},
{
"desc": "distinct mail servers contacted on 25/465/587 (spam)",
"id": "q-smtp",
"key": "src",
"metric": "distinct_dst_ips",
"ports": [
25,
465,
587
],
"threshold": 50,
"window_s": 600
}
]
}
| Field | Meaning |
|---|---|
action | log: when a trigger fires, log quarantine_would_start trigger=<id> src=…
and allow everything. block: quarantine for real. Today: log. |
default_timeout_s | Quarantine length, unless the trigger or manual entry sets timeout_s. |
triggers[].metric |
failed_connects: upstream connect failures (refused, timeout, unreachable, DNS error) in the window.distinct_dst_ips: distinct target IPs in the window (only ports, if set).conns_per_dst: connections to one target IP:port in the window (only ports, if set). |
threshold, window_s | The trigger fires at >= threshold within a sliding window of window_s seconds. |
manual | Quarantines set by the anti-abuse system:
{"id":"qm:…","src":["IP/CIDR"],"user":"…","until":unix,"reason":"…"}. Applied while until is in the future. |
Override: Transit's own admin API (in the fleet app) should list active quarantines and let an operator release or extend one. A release lasts until the end of that quarantine; it does not change the triggers. Exempt Transit's own health checks and internal IPs from all triggers.
Attack ports (VPN/appliance, SMB, RDP, WinRM, Redis/memcached, Docker/k8s/etcd, WebLogic, Elasticsearch, databases,
LDAP/RPC, VNC, IoT/ICS, Hadoop/Consul/Jenkins): 23, 102, 135, 139, 389, 445, 502, 636, 943, 1433, 1883, 2083, 2087, 2323, 2375, 2376, 2379, 3306, 3389, 3443, 4433, 4443, 5432, 5443, 5900, 5985, 5986, 6379, 7001, 7002, 7443, 8009, 8088, 8291, 8443, 8500, 8834, 9200, 9443, 10250, 10443, 11211, 27017, 50000, 60443.
SMTP: 25, 465, 587.