AWAIS — a technical whitepaper.

The shape of the attack surface in vertical SaaS — dealership CRMs, stone fabrication MRPs, landscape ops platforms, dealer BDC stacks — is not the shape of the attack surface in a generic B2B SaaS. The June 2024 CDK Global outage is the canonical example. A single ransomware event at a vendor that sits at the dealer’s operational core froze roughly 15,000 dealerships for weeks, idled hundreds of millions of dollars in daily revenue, and exposed years of customer PII. The CDK incident is not unusual in form. It is unusual only in scale, and the scale is a direct consequence of a vertical-SaaS pattern: one vendor concentrating extremely sensitive data for an entire industry, often with security investment sized for a much smaller risk surface.

The attack surface that AWAIS is built to defend has, at minimum, the following classes of valuable data and entry points:

• •Lead webhooks. ADF/XML and JSON lead intake from third-party shopping sites, OEM-mandated lead vendors, and dealer microsites. Every one of these is a publicly reachable endpoint that accepts free-form customer text and attaches it directly to the dealer’s pipeline.
• •Customer PII. Name, phone, email, address, driver’s license, often SSN for credit applications. Federal Trade Commission Safeguards Rule and GLBA explicitly cover this data. A breach is not just a reputational event; it is a regulatory event.
• •Payment data. Customer deposits, Stripe customer references, ACH details for floor-plan and dealer payout. PCI DSS scope reduction via tokenization is necessary but does not eliminate the payment-fraud attack surface.
• •Vehicle data. Inventory, cost, pack, holdback, dealer-cash incentives. Exposure of cost data is a competitive harm even when no consumer PII is touched.
• •Dealer credentials. Reused passwords, leaked DMS service-account tokens, third-party vendor logins stored as “integration credentials.” The blast radius of one credential-reuse incident is the entire connected pipeline.

The OWASP Top 10 covers the technical vector inventory well. It does not cover the compliance overlay that vertical SaaS operates under: FTC Safeguards Rule (16 CFR Part 314), GLBA financial-privacy obligations, state-level breach notification statutes, and OEM-imposed certification requirements for dealer-connected systems. Defending against SQL injection is necessary. Defending against regulatory non-conformity is also necessary, and that requirement shapes the AWAIS audit trail.

Net: an application detection surface for vertical SaaS has to defend technical attacks, enforce tenant isolation as a first-class concept, and emit an audit trail that survives both internal compliance review and external regulatory discovery. AWAIS is designed against all three.

Gladius operates five sister verticals on a shared foundation: Gladius CRM for auto dealerships, Gladius BDC for dealer business-development centers, GladiusStone for stone fabrication, GladiusTurf for landscape operations, and GoFetch for the consumer vehicle-finder marketplace. Each vertical has its own database, its own tenants, its own users, and its own attack surface. They also share enough architectural DNA that an attack pattern observed against one is, in practice, informative about the others.

Why federate. When a signature lands at one vertical — a credential stuffing run against the BDC sign-in, a probing query against the CRM lead webhook, a tenant-isolation attempt against GladiusStone — the other four are at elevated risk of seeing the same attacker within hours. Single-system detection is too slow against a coordinated probe. Federation lets the mesh defend in minutes what a single vertical would take days to learn alone.

Why CRDT, not consensus. The five verticals are independently deployed, independently operated, and must continue to defend themselves if the federation layer is unreachable. A consensus protocol (Raft, Paxos) would create a hard dependency — a quorum outage would degrade defense at every node simultaneously. A conflict-free replicated data type (CRDT) is the opposite tradeoff: each node accepts local writes, propagates them asynchronously, and merges peer writes deterministically. The defense set is eventually consistent across the mesh and always available locally. We chose monotonic union semantics for the signature set: adding a signature is commutative, associative, and idempotent, so concurrent emissions from multiple verticals merge without coordination.

Federation event shape. Each event carries: the signature hash, the severity bucket (Low / Medium / High / Critical), the origin vertical, the propagation timestamp, and a per-origin signature over the envelope. Signature validation rejects events whose claimed origin does not match the signing key. Replay defense is enforced via the propagation timestamp and a sliding-window idempotency cache.

Steady-state latency. Target propagation from emission at origin to visibility at a peer is under thirty seconds. The public log at /awais/incidents publishes every accepted federation event with its origin and its propagation delta. We expose this number publicly because regression in propagation latency is the single strongest leading indicator of a federation degradation.

Edge to Node bridge. Detection happens at the Vercel edge runtime when possible (low latency, no cold start) and at the Node runtime where Prisma and the federation table live. The bridge is a signed envelope persisted to the federationEvent table that edge runtimes write and node runtimes consume. Cross-runtime trust is anchored by the per-origin signing key, not by runtime identity.

AWAIS is not a replacement for a CDN web application firewall, a SIEM, or a commercial runtime application self-protection (RASP) product. It sits next to all three, owns the slice they don’t own well in vertical SaaS, and inherits the layered model.

The against-no-defense baseline is omitted on purpose. A production vertical SaaS without an application-layer detection surface is not in the comparison set; it is the CDK-pattern incident waiting to happen. AWAIS exists because a CDN WAF and a SIEM, while necessary, are not jointly sufficient to defend a multi-tenant vertical platform.

AWAIS scores itself against a deterministic synthetic adversary on every build. Synthetic adversaries are useful but not sufficient. The system gets sharper when real researchers throw real probes at it. Our public bounty pays $250–$10,000 per accepted finding with a 72-hour triage SLA. A formal $15,000 best-in-show red-team challenge runs on a sandbox tenant. Scope, severity matrix, and rules of engagement are documented at /awais/red-team.