Skip to content

53,600 AI-swarm attacks. Every one stopped in the kernel. 0 bytes leaked.See the evidence

Microseconddefense againstAI swarm, credential theft, IMDS abuse, rogue agentattacks.

Solomon stops AI-agent swarms inside the kernel, investigates every alert with agentic AI and puts a human signature on every critical action.

Every syscall checked in the kernel.

  • AWS
  • EKS
  • Graviton
  • Ubuntu
  • RHEL
  • Amazon Linux
Illustrative

Measured, not marketed. Open numbers.

swarm attempts stopped inside the kernel

Measured · Oct 2026
kernel containment
<µs
of secrets leaked
bytes
agentic investigation
<min

Measured · same swarm, with and without

Same swarm. Nothing leaves.

Without Solomon the swarm emptied the host of its secrets in 17 ms. With Solomon, across 53,600 attempts in four environments, not one byte got out.1

1 Emulated AI-agent swarm evidence run, 1 Oct 2026, and the recorded swarm demo on the same host, Oct 2026. Each bar is the share of a batch of attempts that got secrets out: 20 batches of 100 without, 20 batches of 2,680 with. Canary data only, no real credentials. Method and every number.

With Solomon0 bytesleaked across 53,600 attempts. Every one denied or killed inside the kernel.

100 agents attack. 0 survive.

Replay of a recorded run · Oct 2026 · NVIDIA DGX

Attempts blocked

  1. swarm100 AI agents · 20 rounds · 3 attack vectorst+0
  2. denyIMDS 169.254.169.254 · connect4 -EPERM211 µs p50
  3. killread ~/.aws/credentials · SIGKILLrc=-9
  4. killscrape /proc/self/environ · SIGKILLrc=-9
  5. wave 110 agents · 600 / 600 blocked0 B
  6. wave 250 agents · 3,000 / 3,000 blocked0 B
  7. wave 3100 agents · 6,000 / 6,000 blocked0 B
  8. clientRSS 181.3 MB (+2.0 MB) · 48 containers upok
9,600 of 9,600blocked · with Solomon, 0 bytes out
every secret gonein 17 ms · without Solomon, same host

Replay of a recorded run, Oct 2026: 10, then 50, then 100 concurrent AI agents on an NVIDIA DGX workstation, 20 rounds, 3 attack vectors. The horizontal axis is the order of events in the kernel log, not wall-clock time.

Median time to stop

Stopped in the kernel. Before the read returns.

Each swarm technique is decided on the system call itself. Most never get a byte back.
Technique · MITRE ATT&CKLog scale · 1 µs to 100 ms
  • Steal cloud credentials fileT1552.0016–34 µs
  • Dump process secretsT155215–31 µs
  • Kill the attacker's process treecontainment31–130 µs
  • Steal cloud role credentials (IMDS)T1552.00555–200 µs
  • AI- or analyst-ordered kill, platform → hostend to end4–18 ms
  • Container escape killedT1611 · incl. process start13–56 ms

Medians. Source: Emulated AI-agent swarm evidence run, 1 Oct 2026. The end-to-end kill is an AI- or analyst-ordered kill travelling platform → host; container-escape timing includes process start.

The Solomon loop

Speed without trust is useless. Solomon delivers both.

CISOs won’t let a probabilistic AI touch production on its own. So Solomon splits acting, thinking, investigating and approving into four layers. Each one is a product.
  • OktaIdentity
  • Microsoft Entra IDIdentity
  • CrowdStrike FalconEndpoint
  • SentinelOneEndpoint
  • AWS CloudTrailCloud
  • AWS GuardDutyCloud
  • WizCloud
  • CloudflareNetwork
SolomonOne loop, four layers
  1. 1 · Kernel actsSolomon eBPF Client< 300 µs
  2. 2 · AI agents thinkAgentic AI Platform10,000+ rules
  3. 3 · Agentic investigationSolomon SOC + OmniMesh< 8 min
  4. 4 · Human approvesSolomon XGuard1 tap
native connectors
35
deterministic YARA and Sigma rules
10,000+
tap to approve a critical action
1

The comparison tabs describe what each category typically does out of the box. It is a generalisation, not a claim about any vendor’s product. Connector count from the portal’s integrations catalog today.

Pods · Illustrative9,600 / 9,600 blocked

Every process tree. Its own wall.

Solomon contains each workload in the kernel, before lateral movement.
An agent that reaches for a secret or a namespace is killed where it stands.

in-kernel · eBPF · Intel + ARM

Your own AI agents

Your agents hold real keys. XGuard holds the pen.

Coding, personal, team and ops agents carry live tokens and act with no one watching. XGuard puts a human signature on every critical action.
12
Illustrative

No human gate

Every critical action runs.

  • coding01
  • ops02
  • team03!
  • personal04
  • coding05
  • ops06!
  • team07
  • personal08
  • coding09!
  • ops10
  • team11
  • personal12!
Live tokens
12 exposed
Critical actions
4 running unchecked
Who approved
Nobody

Solomon XGuard

Every critical action waits for a person.

  • coding01
  • ops02
  • team03paused
  • personal04
  • coding05
  • ops06paused
  • team07
  • personal08
  • coding09paused
  • ops10
  • team11
  • personal12paused
Live tokens
Single-use grants
Critical actions
4 paused, 0 decided
Who approved
A person, Face ID

Approval request · XGuard

agent-03 wants to run

kubectl delete deployment payments
Agent
agent-03 · team
Target
prod-eks-1
DLP scan
No secrets in payload
Risk
High · Lateral escalation

Single-use grant · bound to this action, host and window

Explore Solomon XGuard

Measured evidence · 1 Oct 2026

Four environments. 100% stopped.

53,600 swarm attempts on real infrastructure, Intel and ARM, four Linux distros. Every one denied or killed inside the kernel.
Environment1 pill = 800 attempts · stopped
8 AWS servers4 Linux distros × Intel and ARM32,000100%
Kubernetes pods on the EKS node image2,400100%
Live AWS EKS clusterIntel + Graviton · 3 runs9,600100%
In-house production server10 → 100 attackers9,600100%
Total53,600100%

Emulated AI-agent swarm: 50–100 concurrent workers issuing real attacker system calls against canary data (no exploits, no real credentials) on Amazon Linux 2023, RHEL 9 and Ubuntu 20.04/22.04. Lab results from disposable AWS test hosts, a 2-node EKS cluster and an in-house production server. 6 of 8 kill-chain steps are blocked; shell spawn and covert tunnels are monitored. Read the full evidence and method.

Recorded · Oct 2026NVIDIA DGX workstation and a live 2-node AWS EKS cluster (Intel + Graviton). Same swarm, same host: without Solomon every secret is gone in 17 ms; with Solomon, 0 bytes leave, whether 10, 50 or 100 agents attack.

AI economics

Every investigation. 92% cheaper.

Proprietary multi-model routing sends each step of an investigation to the right model, instead of blind calls to the biggest one. Answer quality held at 100%.
Median of 100 tickets
−92%AI cost per investigation

VIntercept proprietary multi-model routing vs blind model calls, median of 100 tickets, answer quality held at 100%.

How Solomon compares.

Solomon compared with typical security categories
SolomonLegacy EDRSIEM + SOARAI SOC copilotMDR service
Kernel-level containmentIn the syscallAfter an alertThrough your EDR
Time to contain< 300 µsMeasuredAfter detectionPlaybook runAnalyst decidesAnalyst decides
Agentic investigation< 8 minPlaybooksBy people
Deterministic rule gate on AI actions10,000+ rulesRules, no AI gate
Human signature on critical actionsHardware-signedApproval stepChat confirmAnalyst call
Protects your own AI agentsXGuard
Works agentless and with a sensorBothSensor onlyAgentless onlyAgentless onlyVaries
Yes With a condition No Best in row

Columns describe what each category typically offers out of the box, not any vendor’s product; individual products vary. Solomon figures are measured (see the evidence page).

Bring Solomon to your infrastructure.

  1. Step 1 of 3
  2. Step 2 of 3
  3. Step 3 of 3

Pilot · step 1 of 3

What do you want to protect first?

Pick one, or press 1 to 4.

A few details

Questions, answered.

All questions
What does VIntercept offer?

The Solomon platform: the Solomon eBPF Client (kernel-level EDR and CDR for Linux), the Solomon Agentic AI Platform (the multi-tenant backend), Solomon SOC (the agentic SOC), Solomon OmniMesh (connectors and remediation), Solomon XGuard (mobile approvals on iOS and Android) and the SolomonPy and SolomonJS SDKs. We also run services on top of it.

How is the eBPF Client different from a traditional EDR?

It decides on the system call itself, inside the Linux kernel. Credential theft is denied before the read completes and an attacker's process tree is killed in place. In our emulated AI-agent swarm evidence run, containment took < 300 µs.

Do I have to replace my SIEM or EDR?

No. OmniMesh connects the identity, EDR, cloud, SIEM and ticketing tools you already own (35 native connectors today), and the agents work on their data.

How do you stop AI agents from doing harm?

Every action an agent proposes first passes a deterministic layer of 10,000+ YARA and Sigma rules and the tenant's autonomy cap, which no alert can exceed. Destructive actions need human approval, and agent identities can propose actions but never approve them.

Where does the eBPF Client run?

On Linux hosts, containers and Kubernetes nodes, including EKS. It is tested on a 14-target kernel matrix on disposable EC2: Amazon Linux 2023 (incl. the EKS AMI) and 2, RHEL 9, Ubuntu 20.04, 22.04 and 24.04, on x86_64 and arm64.

Where do your numbers come from?

Each figure on this site names its source. The security numbers come from our emulated AI-agent swarm evidence run; the method and results are on the evidence page.

Can I try it on my own infrastructure?

Yes. Request a pilot and we will scope a guided pilot on your own workloads with you.

How is my data kept apart from other customers'?

Each tenant gets its own database, event streams, storage, vector memory and secrets. The tenant is taken only from a verified token claim, never from a header or request body.