---
title: "Voidhawk: AI-Native Defense for Web Apps and APIs"
description: "AI-native defense against malicious actions across your apps and APIs: exploits, credential stuffing, account takeover, and abuse of phished credentials. On your existing CDN or WAF."
url: https://voidhawk.ai/
---

# The attackers are AI-native. Your defense should be too.

Voidhawk detects malicious actions across your apps and APIs, explains the intent, and blocks them at your existing CDN or WAF. From vulnerability exploits to credential stuffing, account takeover, and abuse of phished credentials.

Status: built by ZioSec, any CDN or WAF, one click on AWS WAF.

## Inside the Voidhawk console

1. **See every request. Know the source.** 244,293 requests in one day, classified by source and client type: verified crawlers, AI search, AI agents, and 57,873 requests linked to eight attacker groups.

   ![The Voidhawk Traffic & origins view: 244,293 requests in a 24-hour window on a world map of source countries, a traffic chart in 30-minute buckets, the busiest network prefixes, and a client classification table separating verified crawlers, AI search bots and agents, and requests linked to eight named attacker groups.](https://voidhawk.ai/product/console-traffic-origins.webp)

2. **Know what they're after.** One TLS fingerprint links 689 clients to Bitter Inlet. See the evidence for server-side request forgery, the likely next step of metadata credential theft, and all 1,979 stopped attempts. Mark a false positive or escalate.

   ![The Voidhawk actor overview for "Bitter Inlet": 689 clients joined by one JA4 TLS fingerprint, 1,979 findings flagged as server-side request forgery against a URL field, the evidence behind that reading, an AI advisory naming credential theft from a metadata endpoint as the likely next step, all 1,979 findings stopped, and buttons to mark a false positive or escalate.](https://voidhawk.ai/product/console-actor-overview.webp)

3. **See how far one actor reaches.** Track one actor across 689 addresses, 12 networks, and 10 countries through a shared fingerprint. Individual address blocks require review and expire automatically.

   ![Where the "Bitter Inlet" actor came from: a world map and a table of 12 networks across 10 countries, led by South Africa, the United Kingdom, Japan and Australia, with addresses, findings and clients per network, beside the stopped-findings summary and a reviewed option to block a member address.](https://voidhawk.ai/product/console-actor-origins.webp)

Screenshots from the Voidhawk console · example site and addresses.

## Why now: Exploits take hours. Patches take weeks.

AI uncovers decades-old bugs and ships new ones. Exploitation often starts before a patch exists. Voidhawk writes a virtual patch to your WAF while your team builds the fix.

- **12 of 12** January 2026 OpenSSL CVEs found by AI; some dated to 1998
- **44%** of AI-generated code fails security tests
- **1 in 4** exploited CVEs hit by publication day
- **43 days** median patch time, up from 32 last year

Sources: AISLE, Jan 2026 · Veracode GenAI Code Security Report, Jul 2026 · VulnCheck State of Exploitation 1H 2026 · Verizon DBIR 2026 · Mandiant M-Trends 2026.

## CVEs have exploded.

48,185 CVEs in 2025. Up 20% in a year. AI helps researchers find vulnerabilities and attackers exploit them at scale.

- **Patching is a race.** Attackers automate exploitation. Your team still needs a patch window.
- **Your team pays the price.** Critical disclosures turn into emergency patches and lost weekends. The advisories keep coming.
- **Signatures arrive second.** Someone must see a new exploit before writing its signature. That delay leaves you exposed.
- **The agents aren't the problem.** AI agents deliver the exploit. Scale makes it dangerous.

### The disruption: Offense got AI first. Defense must catch up.

Attackers adopt AI faster than defenses adapt. The gap keeps growing.

- **Attackers, super-powered.** In a 2025 campaign disrupted by Anthropic, AI ran 80–90% of a state-sponsored espionage operation against about 30 targets. (Anthropic, november 2025)
- **Frontier models, by accident.** In July 2026, two OpenAI models escaped a cyber evaluation sandbox, partly through a zero-day, and breached Hugging Face's servers seeking test answers. (Openai and hugging face, july 2026)
- **Personal agents, everywhere.** DataDome counted nearly 25 billion AI agent requests in Q3 2026. Agentic browser traffic grew 4.7× in September alone. (Datadome AI traffic report, q3 2026)

**Voidhawk closes the gap.** Voidhawk brings offensive AI to defense. It reads attack intent and stops the exploit at your existing CDN or WAF.

### Case in point: PHP2Shell

**The bug was in PHP itself.**

PHP2Shell names two PHP engine flaws, CVE-2026-109618 and CVE-2026-109619, each rated CVSS 9.2. Unauthenticated requests can run code as the web server worker. Affected runtimes include PHP-FPM, Apache mod_php, php-cgi, FrankenPHP, and LiteSpeed. Researchers demonstrated exploits against WordPress, Drupal, and Joomla.

Researchers found the flaws with AI-assisted tools. Details leaked hours before the fix. Attacks were reported before patches shipped.

You still patch PHP. Voidhawk protects against the exploit while you do; it does not discover or fix zero-days.

If PHP2Shell traffic hit a site running Voidhawk:

- **It reads the intent.** It recognizes a request trying to execute server code, even when the encoding changes.
- **It knows when an attack lands.** It detects traffic using shell access when an exploit succeeds.
- **It writes the virtual patch.** It writes a rule to your WAF to stop the attack before you patch PHP.
- **It alerts your SOC.** Your team sees what was attempted and stopped, as it happens.

## Rules vs reasoning: Signatures scale with headcount. Reasoning scales with compute.

Attackers adapt. Signatures match the past. Voidhawk reasons about the request in front of it.

| | The rule-writing model | Voidhawk |
|---|---|---|
| Who writes the rules | Vendor researchers; one rulebook for everyone | AI reasoning on your traffic, as attacks happen |
| What it covers | Software the vendor has studied | Your apps, APIs, and MCP servers, including bugs without CVEs |
| How it scales | With researchers | With compute |
| What your team sees | A rule ID and a verdict | Evidence, reasoning, and action taken |
| Where it runs | The vendor's own firewall or edge | Your existing CDN or WAF; one click on AWS WAF |

## Keep your CDN and WAF. Add the intelligence.

Any CDN or WAF. No code changes. No new architecture.

- **AWS WAF** (one click): Connect Voidhawk to your existing AWS WAF in one click.
- **Any other CDN or WAF** (guided setup): Our team connects Voidhawk to your existing CDN or WAF.

1. **Connect your edge.** Connect your existing CDN or WAF. No code changes.
2. **Read the intent.** Understand what each request is trying to do.
3. **Write the virtual patch.** Stop the exploit with a rule written to your WAF in real time.
4. **Alert your SOC.** See what was attempted and stopped, as it happens.

## Close the gap between exploit and patch.

Block the exploit. Ship the fix on your schedule. Get a technical walkthrough of Voidhawk on your stack. Now in early access.

## Contact

- Book a technical call: https://calendly.com/aaron-ziosec/15-min-chat
- Early access: info@ziosec.com, or the form at https://voidhawk.ai/contact
- All pages, in Markdown: https://voidhawk.ai/llms.txt