Skip to content

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

Solution · Endpoint and Workload Security

Linux EDR that blocks, not just alerts.

One lightweight agent for Linux hosts, containers and Kubernetes, on Intel and ARM. It decides on the system call itself: credential theft is denied and the attacker’s process tree is gone before the next step runs.

solomon-client · host-17enforce
  1. python3 agent.pyscope ai-worker
  2. └ sh -c 'cat ~/.aws/credentials'
  3. └ openat ~/.aws/credentialsdeny · EPERM
  4. └ python3 worker.py ×50
  5. └ connect 169.254.169.254:80deny · EPERM
cgroup.kill · whole tree gone31–130 µs

Illustration of a blocked process tree. The kill time is the measured median range.

Emulated AI-agent swarm evidence run, 1 Oct 2026
memory, at runtime and at peak under attack1
CPU on our production server; 0 dropped events in 19 hours2
fleet-wide enforce, monitor or kill switch, no restart3
attempts on 8 AWS servers, Intel and ARM, all stopped4
  1. 1Emulated AI-agent swarm evidence run, 1 Oct 2026: agent memory at runtime and at peak while under attack.
  2. 2Emulated AI-agent swarm evidence run, 1 Oct 2026: live production server, 19 hours.
  3. 3Kill switch and live policy reload measured at 0.51–0.60 s on the kernel matrix (Solomon_Client Actions run 36951197586, SCORECARD_PUBLIC.md SC-8).
  4. 4Emulated AI-agent swarm evidence run, 1 Oct 2026: 8 AWS servers · 4 Linux distros × Intel and ARM.

Capabilities

Prevention on every Linux workload.

Most endpoint agents watch and report. The Solomon eBPF Client blocks in the kernel, then tells the Platform what it did.

  • Endpoint detection and response (EDR), with prevention

    • Policy runs as eBPF programs on BPF Linux Security Module (LSM) hooks, so a denied call never completes
    • Where BPF LSM is missing, the process is killed in the kernel before the call returns
    • Covers credential files, cloud metadata, container escape, egress and persistence
  • Cloud workload protection (CWP) for containers and Kubernetes

    • Runs on Kubernetes nodes as a systemd service or a DaemonSet installed with Helm
    • Rules scoped to pods by label, namespace, pod name or service account
    • Container-runtime sockets, escape syscalls and host-file tampering blocked
  • Kernel telemetry in OCSF

    • Process, file and network events straight from the kernel
    • Normalized to the Open Cybersecurity Schema Framework (OCSF) and sent over mTLS
    • Spooled on disk and replayed when the link to the Platform drops
  • Self-protection (anti-tamper)

    • Attempts to kill or trace the agent are refused in the kernel, or the caller is killed
    • Policies stay pinned in the kernel while the agent restarts
    • Deny storms are contained instead of flooding the host
  • YARA scanning and indicators of compromise (IOC)

    • Bundled YARA rules scan files on the host
    • IOCs pushed by the Platform are enforced on the host
    • Executables blocked by file hash in the kernel
  • Response on command

    • Kill an attacker's process tree or isolate a host with in-kernel cgroup filters
    • AI- or analyst-ordered kills reach the host in 4–18 ms end to end
    • Every policy and command is signed by the Platform and verified on the host

Built for production

Light enough to leave on everywhere.

Under attack the agent stayed under 150 MB of memory and used 1.7% CPU on a live production server, with no dropped events. It runs on Intel and AWS Graviton alike, on bare hosts, in containers and on Kubernetes nodes.

After install it only reports. You turn on blocking rule by rule and workload by workload, and a fleet-wide kill switch returns everything to monitor mode in half a second.

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.

Supported todayLinux
Distributions
Amazon Linux 2023 · RHEL 9 · Ubuntu 20.04 / 22.04 / 24.04
CPU architectures
x86_64 (Intel) · arm64 (AWS Graviton)
Kernels
5.10 or newer, tested up to 7.0
Kubernetes
Amazon EKS and k3s nodes and pods
Packaging
GPG-signed .deb / .rpm repositories · Helm chart
Not supported
Windows and macOS

No collateral

Approved tools keep working.

Approvals are tied to the file itself (device and inode), never to a process name, so an attacker cannot rename a tool to slip through.

On Kubernetes nodes the installer approves the kubelet, the CNI and the other control-plane binaries for you. Approved pods and unscoped twins are left alone, and every enforcing rule is tested against that in CI.

exceptions.jsonby file identity
{
  "executables": [
    {
      "name": "backup-agent",
      "path": "/usr/local/bin/backup-agent",
      "rules": ["cred_access"],
      "why": "reads the backup role keys"
    }
  ]
}

Let’s stop the swarm. Together.

Run Solomon against your own workloads in a guided pilot. We’ll show you the kernel log, the investigation and the approval, on your infrastructure.