New · AI agents are already inside bank accounts and real attacks. Read the brief →
SCROLL TO FOLLOW THE ATTACK
Evidence for AI-agent attacks

Every AI agent leaves a trail.

Attribytee is the evidence layer for AI-agent attacks. We rebuild what the agent did, seal it in a tamper-evident certificate, and give insurers the proof they need to pay the claim.

Request accessFor investors
88/88MVP requirements passed215backend tests passing6founders & engineers$16.9Bcyber insurance market

Machines attack.
People file the claim.

1 / 3
01 · Speed
24–48h

Flaws are hit before you can patch.

The average time to exploit a critical vulnerability dropped from 4.76 days to one or two days.

FortiGuard Labs, 2026 Global Threat Landscape Report
02 · Scale
+389%

Ransomware victims, in one year.

7,831 organisations were hit, up from about 1,600. Automation is doing the heavy lifting.

FortiGuard Labs, 2026 Global Threat Landscape Report
03 · Cover
42%

Of companies have AI exclusions.

Insurers are carving AI-driven losses out of cyber policies. When cover is in doubt, the evidence decides.

Delinea survey, via FinTech Global
How it works

Four steps.
One sealed claim.

  1. 01

    Detect

    Your security tools catch the attack and hand the incident to Attribytee.

    EDRSIEMSOCWebhook
  2. 02

    Record

    Every log, file and trace is stored write-once and fingerprinted, so nothing can be swapped later.

    SHA-256write-oncein-region
  3. 03

    Rebuild & attribute

    The attack is rebuilt minute by minute, and the AI model family behind it is named with a confidence level.

    timelinemodel familycompleteness
  4. 04

    Seal & certify

    The case closes as a signed certificate your insurer can verify on its own, and the claim moves.

    hash-chainedsignedverifiable

A market forming right now.

$16.9B

Global cyber insurance premiums in 2025, forecast to reach $30–50B by 2030.

Gallagher Re, via Business Insurance (Jan 2026)
$6.3B

GCC cybersecurity market in 2026, growing to $9.3B by 2034.

IMARC Group
42%

Of companies already have AI exclusions in their cyber cover. Every disputed claim needs evidence.

Delinea, via FinTech Global

Now launching in the Gulf.

Early access for insurers, response firms and banks in Qatar, Saudi Arabia and the UAE. Claims data stays in-country.

Request accessRead the threat brief
Product

From AI-agent attack to sealed evidence.

Attribytee starts where detection ends. Your security tools catch the attack; we turn what happened into a certificate an insurer, adjuster or regulator can check for themselves.

Analyst console

One case, every piece of evidence.

Create a case, upload evidence or stream events by webhook. Every file is stored write-once and addressed by its SHA-256 digest, so nothing can be swapped later.

Reconstruction

The attack, minute by minute.

Events from different sources are normalised into one timeline. Verified facts, source observations and inferred steps stay clearly separated, and partial timelines carry a completeness score.

Certificate

Sealed once. Checkable forever.

The case closes as a signed PDF and JSON certificate. Anyone can verify it later, online or with a standalone tool, without trusting us to be online. Corrections create a new certificate that points to the old one.

Three products. One engine.

Certify

Sealed certificates

Insurer-ready evidence for each AI-agent incident. Graded, signed, portable.

Registry

Cross-victim matching

A privacy-preserving registry that checks whether the same hostile agent hit other organisations.

Connect

Works with your stack

Generic webhook and API, manual upload and reference connectors for the detection tools you already run.

What we do
  • Preserve evidence, untouched
  • Rebuild machine-observable activity
  • Name the AI model or toolkit family, with a confidence level
  • Seal the result for insurers and courts
What we never do
  • Claim a named person ran an attack
  • Hack back or attack anyone's systems
  • Let uploaded content act as instructions
  • Move raw evidence out of its region
Threat brief · October 2026

The 48 hours that changed the claim.

In late October 2026, critical incidents hit nearly every layer of the internet within two days. Several involved AI tools or AI agents directly. This is why evidence now matters as much as defence.

AI agents in the wild

Hacking groups are using AI agents in real attacks.

CrowdStrike researchers found threat groups automating intrusions with AI models, including operations against government and healthcare organisations.

CrowdStrike research ↗

Banks

Unapproved AI agents found inside customer accounts.

Major U.S. banks are detecting third-party AI agents acting in accounts and struggle to tell whether anyone authorised them.

Coverage ↗

Reported in the same window.

PHP2ShellCritical flaws in software behind most websites; Japan's Prime Minister urged site owners to patch the same night.
CloudflareExploit attempts grew from thousands to millions within hours of disclosure.
GitLabCVSS 10.0 flaw allowing unauthenticated takeover of self-hosted servers.
Node.jsHigh-severity remote code execution flaw across current releases.
California schoolsStudent records reported stolen at statewide scale.
RailwayCustomer secrets and API keys possibly exposed; tokens revoked.
LiteLLMAn AI developer library briefly shipped with a password stealer.
CairnCriminals reportedly used an AI hacking tool against 100+ companies.
State Bank of IndiaData breach confirmed.
Medi-CalAbout 970,000 patients' data reported stolen.

Summarised from public reporting and company statements, 21–23 October 2026, including The Hacker News, Cloudflare, CrowdStrike and Axios coverage.

24–48h

Time to exploit a critical flaw · FortiGuard Labs 2026

+389%

Ransomware victims in a year · FortiGuard Labs 2026

7,831

Organisations hit by ransomware · FortiGuard Labs 2026

42%

Companies with AI exclusions in cyber cover · Delinea, via FinTech Global

Who sent the agent?

If you can't answer that, your insurer can't either. That's the gap Attribytee closes.

See how
Company

Built in ten weeks. Built to last.

Attribytee is built by Evon Intelligence with a dedicated engineering team. We started in August 2026 and finished our first two build phases on 2 October 2026.

Progress so far.

88/88

Requirements passed in Phases 1–2

215

Backend tests in Phase 2

18/18

Live end-to-end scenarios passed

2

Build phases complete, zero open in-scope gaps

Timeline.

AUG 2026

The idea and the pivot

We moved from blocking AI attackers to proving what they did, the evidence layer insurers were missing.

19 AUG 2026

Engineering scope v1

Architecture, evidence model and Gulf data-residency rules defined.

24 SEP 2026

MVP specification

Four-person ownership model and vertical-slice build plan, Qatar-first.

2 OCT 2026

Phases 1–2 complete

Forensic foundation and ingest pipeline: write-once evidence, provenance, audit trail, analyst console.

OCT 2026

Public launch

Website, launch film and early-access programme for Gulf insurers, response firms and banks.

The team.

Leadership
AA

Ali Ather

Co-founder

Leads product and engineering at Evon Intelligence.

SS

Salman Sheikh

Co-founder

Leads go-to-market, partnerships and clients.

TS

Taha Saeed

Product & Team Lead

Delivery, evidence ingest and attack reconstruction.

Engineering
AH

Abid Hussain

Technical Lead / Senior Advisor

Architecture, security reviews and technical guidance.

SQ

Syed Qamar Abbas

Front-end Developer

Analyst console, timelines and certificate verification UI.

WW

Wardet Wahaj

Security & QA Tester

Security testing, QA and backend modules.

What we stand on.

Published methods

No black boxes

Model fingerprinting uses published research methods, versioned so every result names the algorithm that produced it.

Prior work

Proven sealing engine

Built on Evon's sealed-report engine: hash-chained reports with evidence classes.

Region

Gulf-first

Raw evidence, cases, certificates and signing keys stay in the country they came from.

References:Pasquini et al., "LLMmap: Fingerprinting for Large Language Models" (2024) ↗CrowdStrike, agentic AI campaigns ↗FinTech Global, AI and cyber insurance ↗
Investors

The evidence layer for the age of AI attacks.

AI agents now carry out attacks on their own. Insurers are excluding AI losses. Attribytee gives every disputed claim the proof it needs, hosted in the Gulf from day one.

Problem

Stopped attacks still become lost claims.

Teams contain AI-agent intrusions, but can't prove what happened or who was behind it. Insurers deny on "unstructured evidence".

Solution

A sealed certificate per incident.

We rebuild the attack, fingerprint the AI model family and toolkit, and hash-chain the result into evidence anyone can verify.

Why now

Exclusions are spreading.

42% of companies already carry AI exclusions. Exploits land in 24–48 hours. The proof gap is growing every quarter.

Where we sit.

Detection companies stop attacks. Evidence platforms serve insurers from the US. Nobody we found combines AI-agent evidence with in-country hosting for the Gulf.

CompanyDetects AI attacksSeals incident evidenceInsurer-ready certificateCross-victim registryGulf in-country hosting
Detection platforms (e.g. Jesta, CrowdStrike)✓————
Insurance evidence platforms (e.g. Spektrum)—✓✓——
Attribyteevia partners✓✓✓✓

Based on public product pages, October 2026.

How we make money.

Per incident

Sealed certificate fee

Charged for each certificate issued on a claim or investigation.

Platform

Annual licence

For insurers and response firms running cases through the console.

Network

Registry access

Cross-victim matching as an add-on that gets more valuable with every case.

Traction and roadmap.

88/88

Phase 1–2 requirements passed

215

Backend tests

18/18

Live end-to-end scenarios

10 wks

From idea to working pipeline

NOW

Forensic foundation and ingest pipeline live

Write-once evidence, provenance, audit trail and analyst console.

Raising our first round.

To take the Qatar MVP live and sign our first design partners. Deck and demo on request.

Request the deckMeet the team
Contact

Request early access.

We're onboarding our first insurers, incident-response firms and banks in Qatar, Saudi Arabia and the UAE. Investors are welcome too.

Talk to the founders.

Tell us who you are and we'll set up a walkthrough of the analyst console and a sample sealed certificate.