Waxell

Defense

Your security stack watches humans.The attackers are machines now.

Attackers now run AI agents that scan, chain exploits and pull data out in hours. Waxell Defense catches the machines: tripwires they can't resist, fingerprints for known attack agents, and one timeline for everything they touch.

Book a defense demo

30 minutes on a live demo tenant. No commitment.

SOC 2 Type II attested · ISO/IEC 27001:2022 certified · HIPAA with a BAA · aligned to the EU AI Act and NIST AI RMF

defense timeline · live
  1. Decoy AWS key from repo config used from an unknown hosttripwire · source attributedalert
  2. Request pattern matches a known autonomous pentest agentattack-agent signature · boundaryflagged
  3. Agent reached for the decoy admin tool on the gatewaytripwire · decoy MCP toolalert
  4. Three signals correlated into one incidentdefense timeline · incident openedincident
Two different jobs
Defense · this pageAttackers using AI against youOutside agents probing your sites, repos, hosts and tools. Waxell plants tripwires, fingerprints attack agents and builds the incident.For the SOC, the CISO and the MSP running security for clients.Book a defense demo
PlatformYour own people and agents using AIThe AI tools on your laptops and the agents your teams build. Waxell finds them, sets policy, and keeps the record.For security, IT and engineering governing internal AI.See the platform
What actually changed
  1. 01

    The whole chain runs on its own

    Agents now do the recon, the scanning, the exploit and the exfiltration without a person at the keyboard.

  2. 02

    Hours, not weeks

    Work that took a crew weeks of manual effort runs over a weekend, at machine speed, against every exposed asset at once.

  3. 03

    Detection was tuned for people

    Alert thresholds, playbooks and behavior baselines assume a human pace. An agent is through them before the first page fires.

The Defense module

Catch the machines, not just the people

  1. 01Catch them

    Tripwires an attacker's agent can't resist

    Agents are thorough. They read every config file, try every key and call every tool. Decoys put that thoroughness to work: nobody legitimate touches them, so when one is used you get an alert with the source, and almost nothing else ever fires it.

    • Decoy credentials and API keys in repos, config and hosts
    • Hidden routes and decoy endpoints in your apps
    • Web decoys on your sites
    • Decoy files on laptops
    • A decoy MCP tool behind your gateway
  2. 02Identify them

    Fingerprints for known attack agents

    Autonomous pentest and attack agents leave behavioral patterns: how they chain tools, how they probe, how they jailbreak. Waxell checks traffic at the boundary against a versioned pack of those patterns and flags a machine for what it is.

    • Versioned signature pack, updated as new agents appear
    • Detective by design: flags and alerts, never silently drops traffic
    • Per-signal confidence, so the SOC knows what to trust
  3. 03Connect the dots

    One timeline, one incident

    A used decoy, a signature hit, a gateway finding and an exposure result usually arrive as four tickets in four tools. Waxell puts every signal on one timeline and correlates the ones that belong together into a single incident.

    • Every source in one view, filterable
    • Correlated incidents instead of alert piles
    • Cooldowns built in, so a noisy hour doesn't become a pager storm
  4. 04Learn from them

    Every documented AI-agent attack, mapped

    417 documented AI-agent incidents from January 2025 to November 2026, 153 with confirmed victims, each mapped to the Waxell control that catches it, or marked plainly as not our layer. You see which attacks you're covered for and which belong to another part of your stack.

    • Sourced from public incident databases, with provenance on every record
    • Honest not-our-layer tags
    • New incidents mapped as they're published
  5. 05Test it

    Attack yourself before someone else does

    Point a real autonomous attack agent at a sandboxed copy of your environment and run it twice: once with your policies off, once with them on. The report shows what the agent reached, what got caught, which policy caught it, and where there's no control at all.

    • Runs in an isolated sandbox with no path to production
    • Ungoverned vs governed, attack by attack
    • Every confirmed gap becomes a regression test
  6. 06Know your surface

    Exposure checks on what attackers see

    Scheduled checks of your external assets against a large, community-maintained template library, so you see the same openings an attacker's agent finds, on the same timeline as everything else.

    • Scheduled, against assets you own
    • Findings land next to tripwire and signature signals
    • One incident when a probe and a decoy hit line up
How it looks at 2am

Two and a half minutes, one incident

An attacker's agent probes, matches a known pattern, grabs a key it should never have found and tries it. Each step is a weak signal on its own. On one timeline it's an incident with a source, a sequence and evidence, before the agent finds anything real.

Illustrative. What lands on the timeline when an attacker's agent works through a company in two and a half minutes.

incident · acme-prod · tonight
  1. 02:14:08Burst of directory probing across three public hostsexposuresignal
  2. 02:14:51Traffic pattern matches an autonomous pentest agentsignatureflagged
  3. 02:16:22Decoy deploy key from a config file used to authenticatetripwirealert
  4. 02:16:40Same source called the decoy admin tool on the MCP gatewaytripwirealert
  5. 02:16:41Four signals correlated, on-call paged with the full chaintimelineincident
red-team run · sandbox · ungoverned vs governed
attackpolicies offpolicies on
  1. Read secrets from a config file and replay themtripwire: decoy keyreachedcaught
  2. Prompt injection in a fetched page to exfiltrate a CRM exportgateway: injection in tool resultreachedblocked
  3. Bulk delete through a database toolapproval on destructive callsreachedheld
  4. Pivot through an unmanaged laptopgap, mapped to EDRreachedno control
The red-team report

See what an attacker's agent would reach

The same autonomous agent, the same sandbox, run twice. The difference between the two columns is what your defenses are worth, attack by attack, with the control that did the work named.

Illustrative report shape. The gap row is the point: you see exactly where another layer has to carry the load.

Built on the Waxell Platform

Defense runs on the same plane that governs your internal AI, so the decoy alert, the gateway finding and the audit trail share one record. You don't need the Platform to run Defense, but if you have both, they connect.

How a rollout goes
  1. Day oneHosted tripwires: decoy keys and web decoys you place yourself. Nothing to install, and the first alert can fire the same afternoon.
  2. Week onePlacements across repos, config, hosts and laptops, a decoy tool on the gateway, and the signature pack switched on for your boundary.
  3. Week twoExposure checks scheduled against your external assets, and the first red-team run against a sandbox, with the ungoverned vs governed report.
  4. OngoingOne timeline for the SOC, incident intelligence mapped as new attacks are published, and every confirmed gap kept as a regression test.

We're onboarding a small group of design partners now. Book a demo and ask for a seat.

What Defense is not
  • Not a replacement for your EDR or SIEM. They defend the traditional perimeter; Waxell catches the AI-agent leg they weren't tuned for, and sends them the incident.
  • Not a promise to stop outages, ransomware or the breaches in the news. Waxell detects, identifies and keeps the evidence.
  • Not the same job as governing your own AI. That's the Waxell Platform, and you can run either one without the other.
Questions security teams ask

Defense watches for attackers who point AI agents at your company: decoys they touch, signatures they match, the incident they leave behind. The Waxell Platform governs the AI your own people and agents use. They share the same alerting and evidence trail, but they answer different questions and you can buy either one.

No. Waxell sits alongside them. Your EDR and SIEM watch hosts, networks and people. Waxell watches for AI agents: the attacker's agent touching a decoy, matching a signature or working through your tools, assembled into one incident you can route to the tools you already run.

Quiet by design. A decoy has no legitimate use, so an alert means something touched it that shouldn't have. Signatures carry a confidence score, alerts have cooldowns, and new detections run in a log-only period before they page anyone.

Not to start. Hosted decoys work on day one with nothing installed. Laptop decoys come with the Waxell endpoint app, and the gateway decoy sits behind the MCP gateway if you use it.

It runs against an isolated sandbox seeded for the test, never your production systems, with a hard time limit, a budget cap and network rules that only reach the sandbox.

Different layer. Waxell doesn't stop infrastructure outages or ransomware. It catches the part of an attack that runs through AI agents, identifies it and keeps the record of what happened.

Yes. Partners run Defense per client from one console, with each client's timeline and incidents kept separate.

Waxell is SOC 2 Type II attested and ISO/IEC 27001:2022 certified, supports HIPAA with a Business Associate Agreement, and is aligned to the EU AI Act and NIST AI RMF.

Watch an attacker's agent trip a decoy, on a live demo tenant, in thirty minutes.

Book a defense demo

30 minutes on a live demo tenant. No commitment.

SOC 2 Type II attested · ISO/IEC 27001:2022 certified · HIPAA with a BAA · aligned to the EU AI Act and NIST AI RMF