New Kerovan for tokenized funds. Free pilot from mid November. See Kerovan →

Valgard · Your own cyber defense team

AI attack tools find open doors in hours. Valgard finds yours first.

Runs on hardware you own. It hunts down your weak spots, blocks them on a timer and proves it for NIS2 and DORA.

Valgard · Your own cyber defense team

AI attack tools find open doors in hours. Valgard finds yours first.

Runs on hardware you own. It hunts down your weak spots, blocks them on a timer and proves it for NIS2 and DORA.

The threat

Attackers already use AI. They start with the systems you check least.

This is not a forecast. The FBI, CISA and Microsoft have all published on it. AI agents now do the slow parts of an attack: finding targets, writing exploit code and trying one path after the next.

FBI FLASH, November 2026

Under three hours

The FBI says attackers use ARTEX, an open-source AI agent framework, to automate intrusions at machine speed, from reconnaissance to data theft. Some went from first target to privileged access in under three hours.

FBI FLASH, November 2026

The forgotten systems first

ARTEX is good at finding and exploiting systems that get less attention: partner and supplier portals, legacy services, third-party integrations. It also lets attackers with little skill run complex campaigns.

Microsoft, October 2026

Hundreds of organizations

A China-state aligned actor used ARTEX to automate attacks on hundreds of organizations. Several times it took less than 3 hours from first target to privileged cloud access.

CISA and FBI, August 2026

AI-written exploit code

Five US agencies warned that attackers use AI to write exploit scripts for exposed Siemens S7 controllers in energy, water and manufacturing. They call it an active threat, not a theoretical one.

CrowdStrike, 2026

AI-enabled attacks up 89%

AI-enabled adversaries increased operations by 89% year over year. Average eCrime breakout time fell to 29 minutes.

M-Trends 2026

Before the patch exists

Mean time to exploit dropped to an estimated minus 7 days. Exploitation routinely starts before a patch is even released.

The FBI's first advice: inventory every internet-facing application, including the peripheral ones, and preserve your sign-in, audit and SIEM records. That is the job Valgard is built for. The answer to AI speed is not panic. It is getting to your own weak spots first, with a person in control. With Valgard, blocks lift on their own timer and bigger changes wait for your people, and every step leaves the signed proof NIS2 and DORA ask for. Figures are context only and do not refer to Scalytics customers or partners.

The counter system

What is Valgard?

What it is

A proactive cyber defense addon. It finds weak spots on your own systems before attackers do, blocks the ones attackers use with a block that lifts itself on a timer, and proves what it did with signed records. Bigger changes wait for your approval.

What your auditor gets

A signed record for every finding: scope version, scan output hash, approvals and rescan verdict. It is evidence for NIS2 Article 21 (vulnerability handling, checks that measures work, asset management), plus report drafts for the NIS2 24 hour and DORA 4 hour clocks.

Who it is for

Security, IT operations and compliance teams at mid-size EU companies that fall under NIS2 or DORA.

The problem it solves

AI makes attacks faster, known flaws get exploited before teams can patch, and auditors want proof of how vulnerabilities were handled.

What it connects to

The Kafka you already run (Confluent, Apache Kafka, Redpanda or Amazon MSK), your F5, Palo Alto and OPNsense firewalls, the firewall, F5, DNS and web logs you already collect, your ticket system, and the CISA KEV and FIRST EPSS feeds.

Where it runs

In your environment, on your hardware and your S3, as an addon to Lascaris Streamhouse or any Kafka installation. Works air-gapped.

Category

Continuous self-scanning, exploited-vulnerability matching, time-boxed blocking and audit evidence for NIS2 and DORA.

The name

Valgard comes from the Old Norse name Valgarðr, a guard's name. It sits next to Kerovan and Helios.

Turn it around

The attackers' playbook, run on your own systems first.

We use the tools the attackers use, to counter them. Valgard reuses the pipeline pattern of AI attack tools such as ARTEX: discovery, scanning, matching against known exploited flaws, then a retest. We reuse the pattern only, not the code. The difference is where it points, what stops it and what it leaves behind: a signed record your auditor can read.

AI attack pipeline (ARTEX) Valgard
Target Any organization it can reach, across many sectors Only the asset list you sign. Out of scope means blocked and logged
Discovery Autonomous recon of overlooked systems such as partner portals and legacy services Grows only inside your signed seeds. New finds wait for a person to approve before any scan
Tools AI agents drive scanners such as nmap, nuclei, httpx and ffuf Safe profile by default, rate-limited, no flooding. Intrusive checks only in an approved window, with a named approver
Speed Chains attack techniques at machine speed When KEV or EPSS changes, matches your inventory at once, then hunts back through months of your own logs in S3
Scope gate ARTEX removed its own authorization-scope mechanism Every task passes the Lascaris policy check against your signed scope
Retest A retest agent with three verdicts: still reproducible, fixed, cannot confirm The same three verdicts, signed into your audit trail
Ends in Privileged access and data theft A time-boxed block, a rescan and signed proof for NIS2 and DORA

Built by veterans of the German armed forces, so chain of command is built in: your scope, your approval, your signed record. The ARTEX column comes from the FBI FLASH and from our own review of the ARTEX source code (AGPL-3.0). Valgard does not ship ARTEX code.

The problem

Attackers move in hours. Close the door before they get there.

Known flaws are still the way in, and mid-size teams are short of hands. Valgard finds the open door and closes it before an attacker walks through. If you fall under NIS2 or DORA, the same work also covers your duty to handle vulnerabilities and report fast.

Exploitation

Up 34% as a way in

Exploitation of vulnerabilities grew 34% as a first step into a breach, with a focus on perimeter devices and VPNs. It now accounts for 20% of breaches.

Mid-size companies

Gaps they did not know about

Exploited vulnerabilities were the top root cause of ransomware for the third year, at 32%. 40.1% of victims were hit through a gap they did not know about, and 39.4% lacked people. Survey of 3,400 firms with 100 to 5,000 staff.

Trust

Automation needs a brake

Only 11% fully trust AI for mission-critical tasks. Valgard does not ask for that trust. Its blocks lift on a timer and bigger changes wait for a person.

NIS2 Article 21

Vulnerability handling is a duty

Measures must include vulnerability handling and disclosure, checks that the measures work, and asset management. Entities that fall short must correct without undue delay.

NIS2 Article 23

24 hours, 72 hours, one month

An early warning within 24 hours, a notification within 72 hours and a final report within one month.

DORA

Four hours for a major incident

The initial notice is due within 4 hours of classifying an incident as major, and at most 24 hours after becoming aware of it.

How it works

Scope, find, hunt back, block, rescan, prove.

Six steps, all in your environment, from your signed scope to KEV and EPSS prioritization to signed proof. Valgard does the checking. Your people make the calls that matter.

01 · Scope

You sign what is yours

You sign the asset list. Every scan is checked against it before it runs. Nothing outside the scope is touched.

02 · Find

Always scanning, safely

Safe scans of your own assets, all the time, with rate limits. Each new KEV entry or EPSS update is matched against what you run right away.

03 · Hunt back

Was it already used?

When a flaw you run lands in KEV, Valgard replays months of your firewall, F5 and web logs from S3 to look for earlier attempts.

04 · Block

Time-boxed, with rollback

IP blocks on F5, Palo Alto or OPNsense run on a timer and roll back on their own. Virtual patches, rule changes and closed ports wait for a person.

05 · Fix and rescan

A ticket, then a retest

A fix ticket with the evidence, the KEV due date and the EPSS score. After the fix, a targeted rescan says fixed, still reproducible or cannot confirm.

06 · Prove

Signed, start to finish

A signed record of scope version, scan output hash, approvals and rescan verdict. Plus NIS2 and DORA report drafts your team reviews and sends.

Red lines

What Valgard will not do.

A defense tool has to be clear about its limits. These are ours.

No hack-back

Valgard is defensive only. It scans and blocks on your own assets and never acts against anyone else's systems.

Nothing outside the scope

Assets hosted by third parties, such as SaaS or a CDN, stay out unless the host has given written permission. Scans never flood.

No 24/7 people

Valgard is software, not a managed service. If you need people around the clock, run it next to your MDR provider.

Your environment

Self-hosted, and air-gapped if you need it.

Valgard is software you run. Your data stays in your environment, and it works air-gapped.

Your data

Logs, findings and evidence stay in your S3, in your environment. It works air-gapped: KEV, EPSS and ET Open rules are plain downloads you can mirror inside.

Your rules

Detection runs on rules you can read: Suricata, Sigma and YARA. Each case cites the rule hits behind it. Nothing goes to an outside cloud service.

Your decisions

You sign the scope. Blocks expire. Bigger changes wait for your people. Every action is signed and logged.

Low friction

If you run Kafka, your security log pipeline is already there.

Valgard runs on the Kafka you already run: Confluent, Apache Kafka, Redpanda or Amazon MSK. Or on Lascaris Streamhouse, our own system for Kafka on your S3. Your firewall, F5, DNS and web logs flow into topics. Valgard reads them as a consumer and writes cases that cite each rule hit and Kafka offset.

Typical AI SOC service Valgard on your Kafka
Where it runs Vendor cloud, often SaaS on AWS Your hardware, your S3, works air-gapped
Who operates it The vendor Your team
Your logs Sent to the vendor or the SIEM Stay in your own S3 bucket
Outside services Hosted AI services such as OpenAI, Anthropic or Google Nothing goes to an outside cloud service

No Kafka yet? Lascaris Streamhouse brings the Kafka engine, on your own S3. Map Valgard onto your stack

The Scalytics stack

Works with the rest of your stack.

Valgard is an addon, like Kerovan and Helios. It runs on the Kafka you already have, and on Lascaris Streamhouse, our own system.

Runs on

Lascaris Streamhouse

Kafka on your own S3, with stateless brokers. Every action passes a policy check first, and every decision lands in a signed, append-only audit trail in your bucket. Valgard uses that check and that trail for every scan and every block.

Kafka Backup

A locked copy of the evidence

Lascaris keeps a locked backup of your Kafka topics on S3 under object lock, on separate credentials, keys and bucket. Valgard keeps a locked copy of its evidence topic there too, under object lock.

Helios

The open energy control plane

Helios is self-hostable, built on standard Kafka transport and air-gappable.

Kerovan

Tokenized fund reconciliation

Kerovan runs in your environment and keeps a signed, hash-chained audit log for DORA and MiFID evidence.

Read more: Lascaris Streamhouse, Kafka Backup, Helios, Kerovan.

White paper

From sensor to superintelligence.

How AI agents work on the Kafka you already run, and how you keep every one of their decisions under control. It covers why Kafka is the right foundation, how Lascaris checks each action against your policy before it runs, and where it runs in your own infrastructure.

  • The four questions to answer before your first agent goes live.
  • Sense, decide, check, record: one agent loop, step by step.
  • The architecture, the deployment options and how to start a pilot.

Compare

How Valgard compares.

We read the public material of the products and services buyers name most, between 14 and 19 November 2026. "Not claimed" means we did not find it in that material. It does not mean the product cannot do it. The Valgard column shows what Valgard is built to do.

Valgard CrowdStrike Falcon Adversary OverWatch Dropzone AI Edge Elastic Security Sekoia Self-Hosted Managed MDR service
Who runs it You. Software only CrowdStrike, 24/7 Dropzone, in your infrastructure You, self-managed You, self-hosted The vendor, not you
Does your data leave your network? No. Stays on your network Yes, to CrowdStrike Cloud No. Stays in your infrastructure No, when self-managed No, when self-hosted Typically yes, cloud-delivered
Response Time-boxed blocks with rollback. Bigger changes need approval Hunters alert and advise Blocks IPs, disables accounts AI Assistant and Attack Discovery Not claimed Not claimed
Self-scan against KEV and EPSS Yes, inside your signed scope Not claimed Not claimed Not claimed Not claimed Not claimed
Retro-hunt in your own S3 Yes Not claimed Not claimed Not claimed Not claimed Not claimed
What your auditor gets Signed evidence. NIS2 and DORA drafts Not claimed Not claimed Not claimed Not claimed Not claimed

Sources: 12, 13, 14, 15, 16. Product names are trademarks of their owners. Something out of date? Tell us and we fix it.

Who it is for

For the teams who answer for the attack surface.

Security, IT operations and compliance leads at mid-size companies under NIS2 or DORA. We start in the EU.

Security and IT

Fix the right flaw first

Of everything you run, see what attackers already exploit (CISA KEV) and what is likely next (FIRST EPSS). Those get a block that lifts on a timer while the fix is ticketed.

Compliance and risk

Evidence for NIS2 and DORA

A signed record for every finding, from scope to rescan, and report drafts for the NIS2 and DORA clocks.

SIEM owners

Works next to your SIEM

Valgard does not replace your SIEM. It reads the alerts and logs you already have and leaves cases and signed evidence.

Who builds it

Built by people who have run high-stakes systems.

Sovereign operations rest on trust. Valgard is built by people who have run systems where every decision had to be accounted for.

Military service

Veterans of the German armed forces

The team are veterans of the German armed forces, served in the Air Force and in ground forces. The experience, accountability and chain of command that sovereign operations require are built in.

Enterprise leadership

Energy, finance and data

The team has held senior and executive roles at E.ON, Allianz, Cloudera, Confluent, Evariant and Google, building and running systems in energy, finance, and data at international scale.

Open foundations

Apache Wayang

Scalytics was founded by the co-creators of Apache Wayang, now an Apache top-level project. A company a co-founder previously built, Ecolytiq, was acquired.

Questions

Common questions.

Is Valgard available today?

Not yet. Valgard is in alpha, in internal testing. The first pilots run with a small group of design partners.

How do I become a design partner?

Register your interest through our contact page and tell us you want to be a design partner. We are picking a small group of mid-size EU companies under NIS2 or DORA. In a pilot, Valgard runs on your hardware, inside a scope you sign, and we measure three things: unknown assets and KEV matches found, time to a draft NIS2 early warning from a test incident, and zero unapproved actions.

Isn't that the same tool attackers use?

It is the same approach, not their code. Valgard reuses the pipeline pattern of AI attack tools such as ARTEX: discovery, scanning, matching against known exploited flaws, and a retest. The difference is where it points and what stops it. It runs only on assets you own and have put in your signed scope, and every task is checked against that scope before it runs. There is no hack-back. IP blocks are time-boxed and roll back on a timer, bigger changes wait for a person, and every action is signed and logged.

How should we prioritize vulnerabilities with CISA KEV and EPSS?

Valgard ranks by what attackers actually use. A match with the CISA Known Exploited Vulnerabilities catalog comes first, then FIRST EPSS, the likelihood that a flaw gets exploited next. Those get a block that lifts on a timer while the fix is ticketed. KEV and EPSS are plain downloads you can mirror inside an air-gapped site.

Is Valgard an autonomous pentesting tool?

No. It reuses the pipeline pattern of autonomous penetration testing tools such as ARTEX, which the FBI described in its November 2026 FLASH, but it points only at assets in your signed scope. Safe checks run by default and are rate-limited. Intrusive checks run only in an approved window, with a named approver. Every run ends in a time-boxed block, a rescan and a signed record.

How do we protect F5 and Palo Alto edge devices from known exploited vulnerabilities?

Edge devices are a favourite way in: Verizon found that exploitation of vulnerabilities grew 34% as a first step into a breach, with a focus on perimeter devices and VPNs. Valgard checks your edge devices and other assets against CISA KEV. While the fix is ticketed, it can set a time-boxed IP block on F5, Palo Alto or OPNsense that rolls back on a timer. Virtual patches and rule changes wait for your approval.

What are the NIS2 and DORA incident reporting deadlines?

NIS2 Article 23: an early warning within 24 hours, a notification within 72 hours and a final report within one month. DORA: the initial notice within 4 hours of classifying an incident as major, and at most 24 hours after becoming aware of it. Valgard drafts the reports from its signed records for your team to review and send.

Does Valgard attack anyone?

No. It only scans and blocks on assets you own and have put in the signed scope. There is no hack-back. The signed scope is your written authorisation.

Do we need Lascaris Streamhouse to run Valgard?

No. Valgard is an addon for Lascaris Streamhouse or the Kafka you already run, such as Confluent, Apache Kafka, Redpanda or Amazon MSK.

Does our data leave our environment?

No. Valgard runs on your hardware, keeps logs and evidence in your S3 and sends nothing to an outside cloud service. It works air-gapped. KEV, EPSS and ET Open rules are plain downloads you can mirror inside.

What can Valgard change on its own?

Only time-boxed IP blocks on F5, Palo Alto or OPNsense, and each one rolls back on a timer. Virtual patches, rule changes and closed ports wait for a person to approve. Every action is written to the signed audit trail.

Who decides what is an attack?

Rules you can read: Suricata or Snort signatures on network data, Sigma on logs and YARA on files. Each case cites the rule hits behind it. Bigger actions wait for your people.

Does it replace our SIEM?

No. Valgard reads the alerts and logs you already have, does the checking and the gated response, and leaves cases and signed evidence.

How does Valgard help with NIS2 and DORA?

NIS2 Article 21 asks for vulnerability handling, checks that measures work and asset management. Valgard keeps a signed record from scope to rescan for every finding, and drafts the reports for your team to review and send.

What does it cost?

Valgard is in alpha, so it has no price yet. We will set it with our first design partners. Prices for our live products are on the pricing page.

Valgard · Design partners · For Lascaris Streamhouse or any Kafka

Find it, block it, prove it. On your systems, with us.

We are picking a small group of design partners: mid-size EU companies under NIS2 or DORA. Tell us you want to be a design partner and register your interest. A pilot measures three things: unknown assets and KEV matches found, time to a draft NIS2 early warning from a test incident, and zero unapproved actions.

Pilot

Inside a scope you sign

Valgard checks only the assets you list and sign. Blocks lift on a timer and bigger changes wait for your people.

Setup

On the Kafka you have

Valgard reads the logs already in your Kafka or in Lascaris Streamhouse, in your environment. No data leaves.