66 rules. 261 attack patterns. Read it. Stress-test us.
66
Diagnostic rules
261
Attack patterns
53
Error signatures
145
AWAIS IQ score
Eight rule-category families
Generic descriptions on the public surface — specific rule IDs, weights, and thresholds are shared with security partners under NDA. Every family below ships in the same in-process engine that serves the app.
01
RULE_FAMILY_SQLI
SQL-injection detection across query string, JSON body, headers, and cookie surfaces. Catches union, boolean-blind, time-blind, stacked queries, and out-of-band variants. Tuned to keep false-positive cost below the cost of one analyst minute.
02
RULE_FAMILY_XSS
Reflected, stored, and DOM-based XSS detection. Watches reflected echo of attacker input, stored payloads on write paths, and DOM sinks invoked from window.location and document.referrer. Pairs with CSP report-uri telemetry.
03
RULE_FAMILY_SSRF_RCE
Server-side request forgery to internal metadata endpoints, link-local ranges, and cloud IMDS surfaces. Remote-code-execution detection for shell metacharacters, deserialization gadgets, and prototype-pollution chains.
04
RULE_FAMILY_AUTH_BYPASS
JWT tampering (alg=none, key-confusion, expired-but-reused), session fixation, credential stuffing, and OAuth state-parameter abuse. Cross-references the cross-session identity layer to fingerprint repeat offenders.
05
RULE_FAMILY_RATE_ABUSE
Scraper detection, credential-spray, account-enumeration, and inventory-probe patterns. Distinguishes a noisy partner integration from a hostile crawler by behavior, not just IP volume.
06
RULE_FAMILY_TENANT_BOUNDARY
Cross-tenant data-access attempts: dealerId mismatch on tRPC inputs, foreign-key escapes in raw SQL, and authorization bypass on multi-tenant routes. The category that exists only because Gladius is multi-tenant by design.
07
RULE_FAMILY_PII_LEAK
Compliance-grade enforcement against accidental PII egress: SSN, driver-license, EIN, and full-PAN patterns in response bodies. Aligned to FTC Safeguards plus state DMV data rules.
08
RULE_FAMILY_MESH_SIG
Signature validation on cross-vertical federation messages. When CRM learns an attacker fingerprint, BDC + Turf + Stone + Detail receive a signed propagation event. This category protects the propagation channel itself.
How AWAIS learns
AWAIS learns from three streams, fused into one memory.
The first is the CVE feed — CISA KEV, NVD, and vendor advisories ingested on a 6-hour cadence and projected into the SynthesizedDefenseRule table. That feed is where new attacker techniques become detection rules without a human in the loop.
The second is production traffic across the 5-vertical federation — every request CRM, BDC, Turf, Stone, and Detail serves passes through the same edge instrumentation. Real attacker behavior tunes the false-positive baseline per endpoint, per vertical, per hour.
The third is cross-vertical federation: when a pattern fires in one vertical and the operator confirms it, a signed propagation event hardens the other four in under 30 seconds. That is how a phishing pattern caught in BDC becomes a CRM detection rule before the next request lands.
What we don’t claim
Three claims AWAIS is not making. Publishing the limits is part of the contract.
AWAIS is not a SIEM.
We do not warehouse arbitrary log streams or replace Splunk / Sumo / Elastic. AWAIS detects, decides, and acts at the application edge — it does not aim to be the system of record for every event in your environment.
AWAIS has not been externally pen-tested yet.
The IQ score is externally benchmarked via a deterministic synthetic adversary. A third-party human red team has not yet engaged. Public bounty is open at $250–$10k per finding (see /security) and a free invited red-team week is on the table for qualified firms.
AWAIS does not replace a CDN WAF.
Cloudflare / AWS WAF sit at the perimeter and absorb volumetric layer-3/4 attacks AWAIS will not see. We operate one layer in — at the application edge — where intent is visible and tenant boundaries matter. The two layers are complementary, not substitutable.
For security teams
We run a public bounty against AWAIS itself and the surfaces it defends. $250–$10k per accepted finding. 72-hour triage SLA. No legal threats for good-faith research.
Report a vulnerability
PGP key + scope + payout matrix on /security. Median triage time so far: under 24 hours.
Invited red-team week
Qualified firms get a free, scoped, 5-day engagement against a sandbox of the live federation. Findings count toward bounty. We publish the executive summary if you let us.
AWAIS · Gladius Technologies LLC · 2026