# WAF for _the Agentic Era_

**Virtual Patch web apps in minutes**

Legacy WAFs weren’t designed to keep up with Agentic attacks. Impart allows security teams to respond quickly to zero days and agentic attacks with WAF that can be designed, tested, and deployed in minutes with AI security agents.

## What it Does

### WAF at runtime. _Built for execution._ Not inspection.

AI agents and agentic workflows have fundamentally changed how attacks execute. Impart runtime enforcement controls what requests do, at the speed of AI.

### Inspect

- every request with full context—headers, payloads, session state, and prior behavior

### Decide

- what the request is allowed to do, based on intent, not pattern matching

### Enforce

- before execution completes, blocking injection, exfiltration, and abuse before your app processes the input

## _The AI attack surface is different._ Your WAF has to be too.

AI-driven enforcement with Impart runs across payloads, identity, and session rather than signature matching on individual requests. This is how we’re different.

- AI-driven detection
- Sequence-aware enforcement
- OWASP Top 10 enforced at runtime — injection, XSS, CSRF, and beyond
- Prompt injection and novel AI attack variants detected without prior signatures
- Behavioral intent modeled across every request, not matched against static rules
- Proposes and manages rules from observed behavior — no manual tuning
- Requests evaluated in session context, not isolation
- Multi-step attacks visible before individual requests cross thresholds
- Valid requests prevented from producing invalid outcomes across sessions
- Decisions based on what a session is doing, not just what a single request contains

## Agentic attacks unfold across sessions. _Impart enforces the entire sequence, at runtime._

AI attacks unfold across sequences, not single requests. Impart enforces the entire interaction, at runtime.

### Track

- request chains across sessions, identities, and endpoints
### Prevent

- agent workflows from producing unauthorized state changes
### Stop

- multi-step attacks — reconnaissance, probing, exploit — before they complete
### Enforce

- across the entire interaction, not just the request that crosses a threshold

## WAFs detect. _Impart controls what requests do._

| Capability                | Enforcement position       | Detection model                      | Rule management           | Sequence context         | Novel attack coverage         | Cross-surface context                |
|--------------------------|--------------------------|-------------------------------------|--------------------------|--------------------------|-------------------------------|--------------------------------------|
| Cloudflare WAF          | Edge proxy               | Signatures and managed rules        | Manual tuning required    | Per-request only         | Requires new signatures        | WAF only                             |
| AWS WAF                 | Edge only                | Managed rule sets                   | Manual or managed        | Per-request only         | Requires rule updates          | WAF only                             |
| **Impart**              | Inline, application layer | AI behavioral intent modeling      | AI-proposed, auto-managed | Full session, cross-request | Behavioral, no signature needed | WAF + API + Bot + LLM                 |

## The Impart Approach

### _Allow. Block. Modify._ At the point of execution.

Step 1  
Request intercepted inline at the application layer — before your app processes it

Step 2  
Behavioral analysis evaluates every signal against full session history — payloads, identity, and prior requests — against 200+ detection tags

Step 3  
Decision applied before execution — allow, block, or modify at the point of enforcement

## _One runtime engine._  Every request — web, API, and agentic.

```plaintext
API external GET /api/v2/payments service internal agent AI-driven POST /svc/auth/verify
Position
Inline across web apps, APIs, and agent-driven workflows
device:: known fingerprint
location:: Houston TX
rate:: normal
endpoint:: /users · always
timing:: business hours
device:: new fingerprint
location:: Paris FR
rate:: 47/min
endpoint:: /pays · first
timing:: 02:14 AM
BLOCK if request.headers.match(agent_fingerprint_db) AND request.path ~ /export|dump|bulk → block
```

Intelligence
- AI behavioral model built from your traffic, updated continuously

Timing
- Enforces synchronously — no async gap between detection and block

Live RULE
```plaintext
if normalize(request.body).contains_sqli() → block
```

Autonomy
- Validated rules enforced after simulation, without waiting for a human review cycle

## What a _blocked attack_ looks like in Impart.

Unified coverage. Reliable enforcement.

### Recon

The entity is identified and fingerprinted at first contact.  
Behavioral context begins building immediately across every service.

### Probing

Injection and fuzzing attempts detected as a behavioral pattern, not individual events.
Request velocity and structural deviation tracked across the session in real time

### Exploit Attempt

Intent confirmed across the behavioral sequence.
Payload and session context evaluated together — not matched in isolation.

### Enforcement

Rules generated and validated based on observed behavior — blocking the attacker going forward.
What started as a sequence of valid requests becomes a denial.

## FAQ

### What is the difference between shift left and runtime security?
Shift left moves security earlier in the development lifecycle, focusing on finding vulnerabilities before code ships. Runtime security operates after deployment, inline in the path of live traffic, detecting and blocking threats as they happen. A runtime protection platform is the layer that catches what shift left cannot: threats that have no pre-deployment signature, behave like legitimate traffic, and complete in milliseconds.

### How do you stop AI agent attacks?
AI agents pursue goals across sessions, probe multiple surfaces simultaneously, and adapt continuously. Stopping them requires inline enforcement at the origin, behavioral detection that models intent across sessions rather than matching signatures, and a shared data layer across every surface so a single agent cannot get a clean slate by switching attack vectors. That is what a runtime protection platform is built for.

### What replaces a WAF in the AI era?
A runtime protection platform replaces a WAF by combining behavioral detection, inline enforcement, and shared context across Web Apps, APIs, AI Apps, and LLMs on one data model. Impart replaces the detection model entirely, not just the interface.

### What is runtime enforcement?
Runtime enforcement is the ability to detect and block a threat at the moment the request is made, inline in the path of live traffic, before it reaches your application. It is distinct from detection-only tools that observe traffic and alert after the fact, and from shift-left tools that look for vulnerabilities before deployment.