Backed by Y Combinator · S26

AI that detects and fixes OS vulnerabilities — across every endpoint, in minutes.

Palisade watches your systems at the kernel — syscalls, eBPF traces, process and memory events, live network packets — finds the vulnerabilities other tools miss, and applies the fix fleet-wide. Privilege escalation, remote code execution, memory corruption: caught before the exploit lands.

Palisade CONTROL PLANE
audit-only · 37 rules · 5 packs · 3.6k events · 0 denies
MODEAUDITwatching only — nothing blocked
RULES37policy map · 5 packs
DENIES0only while enforcing
NET-004 MEDIUM Cloud metadata service reachable by every process

Any local process can query 169.254.169.254 and harvest instance credentials — the same path behind the Capital One breach.

MAINTAINER FIX → block outbound to 169.254.169.254 except cloud-init
LIVE FEED — every row is a kernel decision on a syscall
NET-004 fixed → rule installed

Simulation of the live product — same rules, same syscalls. Watch it run against a real kernel →

THE GAP

Modern attacks target the operating system.
Most security tools are watching everything else.

Network, cloud, and application-layer products can't see what happens inside the kernel. That leaves OS vulnerabilities among the hardest problems to detect — and the slowest to fix. Even after you find one, rolling a mitigation out across hundreds or thousands of machines takes weeks.

PRIVILEGE ESCALATION

From user to root

A single local privesc turns any foothold into full control of the host — and of everything scheduled on it.

REMOTE CODE EXECUTION

Payloads out of nowhere

Unknown binaries executing from /tmp is the classic opening move. By the time it's in your SIEM, it already ran.

MEMORY CORRUPTION

Invisible from outside

Use-after-free and heap bugs live below every agent that only reads logs. Detection requires kernel-level visibility.

HOW IT WORKS

Deploys alongside your OS. No rip-and-replace.

Palisade ships as a module next to your existing operating system and attaches at the LSM-BPF layer. AI runs directly on kernel data — then every fix becomes an ordinary, inspectable rule.

01

See the layer nothing else can

Syscalls, eBPF traces, process and memory events, live network packets — streamed from the kernel in real time. AI finds the vulnerabilities monitoring tools miss.

02

AI finds it. You approve the fix.

The host audit ranks real findings by severity. Every finding ships a vetted maintainer fix — nothing is changed behind your back. One click installs it as a kernel rule.

03

Flip to enforcing. The attack dies at the first syscall.

Matched operations return -EPERM from the kernel itself. The payload never executes; the connection is never made. A rollout that took weeks lands fleet-wide in minutes.

AI-NATIVE POLICY

Write policy in English. Enforce it in the kernel.

Describe the rule the way you'd say it in a security review. Palisade compiles it into a kernel-enforced BPF policy — and rejects deny-all footguns before they ever reach the kernel.

AUTHOR A RULE IN ENGLISH
Block reads of /root/.ssh/ except by sshd, ssh, scp
rule 38 file_open deny path /root/.ssh/ except sshd, ssh, scp ✓ installed

Prefer determinism? Compliance packs bypass the model entirely — every rule is preset, reproducible JSON:

cis-linux-level-1 10 rules iso-27001-baseline 7 rules nist-800-53-moderate 6 rules pci-dss-v4 6 rules intrusion-detection 8 rules

LIVE AGAINST A REAL KERNEL

Not a mockup. Every row is a real kernel decision.

Captured on a production Linux host running the kernel-attached daemon (LSM-BPF). Every number, deny, and event below is real syscall data.

The Palisade control plane: audit-only mode with 37 rules, 5 packs and 3.6k events; a live feed of file_open ALLOW decisions streaming from the kernel.
The control plane, live — the feed at the bottom streams real syscall decisions, polled from the daemon over the same Unix socket palisadectl uses.
The Palisade host audit: findings ranked by severity, including sshd permits root login (high) and cloud metadata service reachable by every process (medium) with a one-click Apply fix button.
The host audit — real findings ranked by severity, each shipping a vetted one-click fix. The bottom one is the cloud-metadata exposure: the same path behind the Capital One breach.
BEFORE — AUDIT ONLY An attacker's terminal in audit mode: the payload in /tmp runs and prints 'pwned', and the cloud metadata endpoint answers with HTTP 200.
The attacker's terminal: the payload runs, the metadata service answers HTTP 200.
AFTER — ENFORCING The same terminal with Palisade enforcing: the payload is blocked with exit code 126 and the metadata connection is refused — not one credential leaves the machine.
Same host, enforcing: payload blocked (exit 126), connection never made. Not one credential leaves the machine.

NOT TOY EXAMPLES

Real CVEs, detected and fixed.

Every one of these is a path to root — caught and mitigated before an exploit lands.

CVE-2024-1086

Linux kernel netfilter nf_tables

Use-after-free in the kernel's netfilter subsystem — local privilege escalation to root.

→ root · mitigated
CVE-2025-6019

libblockdev / udisks

Privilege escalation through the storage stack to full root.

→ root · mitigated
CVE-2025-32463

sudo --chroot

Local privilege escalation to root through sudo's chroot option.

→ root · mitigated

WHO IT'S FOR

One kernel vulnerability, multiplied by your whole fleet.

GPU clusters

Training and inference fleets where a single compromised node can reach every job, checkpoint, and credential on the fabric.

Cloud device fleets

Hundreds of near-identical instances: the same vulnerability everywhere, the same metadata service to steal from — and one rollout to fix it.

On-device deployments

Endpoints in the field you can't easily re-image. Palisade mitigates at the kernel until you can patch on your schedule.

WHY THIS EXISTS

Built at the intersection of machine learning and operating systems.

"In red-team and blue-team work, one pattern kept showing up: OS-level vulnerabilities were consistently the hardest to handle. Detection needed low-level visibility most tools don't have — and even after you found one, rolling the mitigation out across hundreds or thousands of machines took weeks. I was convinced this didn't have to be a manual process. So I built Palisade."

Prince — Founder, Palisade prince@palisade-ai.com
  • Red-team & blue-team security work at startups
  • ML & applied-math research at Caltech labs
  • SANS Institute GIAC advisory board — one of its youngest members
  • Top 0.1% in national olympiads
  • Built an operating system from scratch

A fix that used to take weeks.
Palisade applies it in minutes.