Malicious versions of a build system went after the configuration of AI coding tools. 2,349 secrets were taken.
AI agents are a target.Source: GitGuardianWe are building the security infrastructure for how software is made.
Every package installed, every pipeline run, every registry download and every action an AI coding agent takes passes one checkpoint that decides, before anything runs, whether it is safe. Cordon is building all of it, in the open. This page shows the whole platform, what it does for your organisation, and how much of it is already built.
Malicious code stopped at the door, wherever the door is.
One platform, every door, every ecosystem.
Supply-chain security, rebuilt for how software is made today.
- 1An attacker publishes a malicious release, often under a trusted name.
- 2Cordon sees it as it is published and works out what it does, from its code.
- 3The verdict reaches every checkpoint, signed, within minutes.
- 4A developer's install, a pipeline and an AI agent are each refused, with the reason.
- 5Your security team is alerted in its own tools, with every place the package was found.
In 2025, supply-chain attacks became fast, automatic and aimed at AI agents.
A maintainer's account was taken over and malicious versions of packages used everywhere were published.
A trusted name is not enough.Source: WizA self-replicating worm stole publishing tokens and republished itself into more than 500 npm packages.
Faster than people respond.Source: CISA alert, 23 SeptemberEighteen minutes from publish to owned.
A typosquat lands on npm. Cordon does not wait for your next scan: new intel is matched against what every repository, image and cache already contains.
- 13:51[email protected] is publishedOne letter from event-stream, by an account created that morning.
- 14:02Intel flags itIts install script fetches a payload and runs it. Read from the tarball; nothing executed.
- 14:03Matched against every inventoryTwo repositories and an Artifactory cache, found without a rescan.
- 14:03Refused at the firewallFour installs today never landed. One CI job that bypassed the firewall is named.
- 14:09Owned and ticketedOwner from CODEOWNERS, PagerDuty paged, PAY-412 opened in Jira, 48-hour SLA running.
Everyone between a developer's laptop and production.
- Install as usual; malware is stopped first
- Clear reasons, never a silent block
- Exceptions in minutes, not tickets
- Every door covered by one policy
- "Are we affected?" in seconds
- Alerts where you already work
- One registry checkpoint for the whole organisation
- Pipelines that fail on malicious changes
- Works offline and air-gapped
- Agents that ask before they act
- Poisoned tools and skills refused
- Every agent and server inventoried
- A signed record of every decision
- SBOM and VEX ready for the Cyber Resilience Act
- Who approved what, and when
- Supply-chain risk you can see and measure
- No vendor holding your source code
- Accuracy published with its method
Twenty-six parts. Three layers. Built in the open.
Checkpoints at every door; a place where your organisation governs; and intelligence that keeps every checkpoint current. Built means its code and tests pass; Cordon Cloud's launch is under way.
- 1Every new release on the public registries, and every new advisory, reaches Cordon as it is published.
- 2Cordon works out what each release does, from its code, and people confirm before anything is condemned for everyone.
- 3Signed protection goes out to every checkpoint your organisation runs, and works even offline.
- 4A malicious package is refused at the door: on a laptop, in a pipeline, at the registry or by an AI agent, before it runs.
- 5Only minimal records reach your Cordon Cloud: alerts go to your tools, the inventory knows where everything runs, and every decision is kept as evidence.
Every part, and what it does for you
Choose any part to see how it protects your organisation and whether it is built yet.
One checkpoint for every package your organisation downloads.
- A malicious release never reaches a laptop, a CI runner or a build server
- A newly discovered threat is refused everywhere at once
- Developers install exactly as they do today
From "are we affected?" to blocked everywhere, in minutes.
- 1Cordon learns that a package already in use is malicious.
- 2The inventory finds every repository, pipeline, laptop and cluster that uses it.
- 3The owner of each place is alerted, and tickets are opened.
- 4The package is refused everywhere at once, and the evidence is kept.
Enforced by its tests, everywhere it runs.
The console your security team works in.
One list for the whole organisation: what was found, where, who owns it, and how long they have.
The rule, the file, the line, what the code does and what to do next. No silent blocks.
When a package turns malicious, see every repository, pipeline and machine that uses it.
Packages, images, AI agents, MCP servers and skills, on every repository and device.
One checkpoint for every package your organisation downloads.
Every model, agent, tool and skill your software depends on, in one bill of materials.
Every repository, pipeline and device, and which ones still need a checkpoint.
SBOMs, VEX and a signed record of every decision, ready for a regulator.
Slack, Teams, PagerDuty, Jira, ServiceNow, Splunk and more, with nothing lost if a channel breaks.
Plain questions about your supply chain, answered from your own data.
No new place to look. Cordon works in the tools you already use.
It stops a malicious package at the door and tells the right people where they already are, when it matters. Point at any tool to see what your team sees there.
Your agents ask Cordon before they install, add a tool or follow a skill.
Claude Code
Cursor
GitHub Copilot
Codex
Windsurf
Gemini
Cline
Roo Code
Kiro
Continue
opencode
Augment
Zed
Trae
Junie
Goose
Amazon Q
- 1The agent is about to install a package, add an MCP server or load a skill.
- 2It asks Cordon first, and is allowed, or refused with a reason it can read.
- 3Anything it installs still passes your organisation's registry firewall.
- 4Each device reports which agents, servers and skills it runs, so nothing is hidden.
Malware is judged by what it does, never by its name.
Everything that can run in your build or your agent.
- detection rules
- 1,501
- Agent Threat Rules
- 815
- package ecosystems
- 28
Dependencies
28 ecosystems and the lockfiles they write, Bun and Deno included, from npm and PyPI to Conan, Hex, CRAN and Bazel.
- Malware by behaviour in Python, JavaScript, Ruby, PHP, Go, Rust, Java, .NET, Swift, Kotlin and twelve more languages
- Known-malicious releases, and version ranges that would install one
- CVEs with reachability down to the vulnerable function, CISA KEV and EU EUVD marked
- Typosquats, combosquats, lookalike scopes and AI-hallucinated names
- Dependency confusion against your internal namespaces
Builds and pipelines
GitHub Actions, GitLab, Jenkins, Azure, CircleCI, Bitbucket and Buildkite.
- Workflows that run a stranger's code with your secrets (pull_request_target, workflow_run)
- Script injection from issue titles, branch names and commit messages, in every CI dialect
- Secrets sent off the runner, poisoned caches and artefacts
- Compromised and unpinned actions and reusable workflows
- Makefiles, CMake, MSBuild, Gradle and build.rs that fetch and run code
Agents and MCP
Every coding agent in use, 49 of 49 configuration locations, read without starting one.
- 815 Agent Threat Rules, and prompt injection in eleven languages
- Hooks and MCP launches that fetch and run, exfiltrate or open a shell
- MCP servers handed the Docker socket, host root or a plain-http remote
- Poisoned tool descriptions, from local and remote servers alike
- Remote MCP servers read as they serve now, and any tool changed since you approved it
Artefacts
Whatever arrives through the supply chain, opened and read, never run.
- Archives inside archives, safely, with bombs and path escapes refused
- Binaries by what they import, with YARA and ClamAV hand-off
- Model files that execute on load: pickles, PyTorch, Keras
- Models loaded with trust_remote_code, or with a loader's safety switch turned off
- Office macros, remote templates, PDF launch actions
Secrets
60 credential types, anchored to each issuer's real format.
- In the tree, the index and the whole git history
- Checked live with the issuer, only when you ask
- A secret plus egress reported as theft, not a leak
- Reported by hash, never by value
Infrastructure
862 policies generated from the providers' own schemas, plus hand-written rules.
- Terraform, CloudFormation, Kubernetes, Helm, Compose and Ansible
- Images a Kubernetes workload runs: pinned by tag, or with no signature or provenance
- Missing encryption, public exposure, deletion protection, weak TLS
- Found by content, wherever a manifest sits
- Reported as posture, so it informs without breaking builds
Wherever code enters, the same checkpoint.
npm
PyPI
crates.io
Go
Maven
Gradle
NuGet
RubyGems
Hex
Packagist
pub.dev
CRAN
Conda
Conan
vcpkg
CocoaPods
Swift
Hackage
opam
Julia
Nix
Homebrew
Bazel
Helm
Terraform
Ansible Galaxy
Container images
GitHub Actions
VS Code, OpenVSX
MCP servers
Registries delete malware within hours. Cordon is watching before they do.
A security tool in front of every install must never become the attack.
- 1Your source code, file paths, the secrets a scan finds and the details of every finding stay on your machines.
- 2Signed protection and your organisation's policy come to every checkpoint, verified there.
- 3Results go to your own Cordon console only if you choose, as minimal records.
Real malware, real software, and an interval on every rate.
Every number names its set, its size and the date it was measured. Hover a row for its counts and interval.
- 01Real software onlyPublished packages and real malware; nothing made up.
- 02Counted onceOne file under many names is one unit.
- 03Held back, scored onceA set kept out of all design work, scored a single time.
- 04Offline and sealedEvery measurement runs with no network.
- 05Two readers per labelA label changes only when two people agree.
- 06An interval on every rateEach number carries its 95% interval.
What landed, newest first.
- Alerts that are never lost: when Slack, a webhook or any destination breaks, alerts wait and go out when it works again
- Alert routing by ecosystem, quiet hours, and escalation when nobody acknowledges; a second person to approve a new destination
- Email alerts with confirmation and one-click unsubscribe; ServiceNow; webhooks in your own format
- Signed verdicts any scanner can verify offline, and a way to withdraw a wrong one everywhere
- Each laptop's AI agents and tools reported, signed; a warning when a device runs something not approved
- The isolated analysis service's foundations: an environment with no way out, for running new packages safely
- Registry firewall connections that authenticate both ends
- A Free plan with clear limits
- Device names kept private as stable pseudonyms
- Policy simulation: see what a rule would block before switching it on
- One decision API for every checkpoint, returning signed, stored decisions
- Exceptions bound to one package version and an end date, signed when approved
- A tamper-evident record of every decision
- AI agents per MCP server, each with an owner
- cordon-scanner 0.7.0 released on PyPI
- A public malware database, one page per advisory
From a scanner to the default security layer for software.
- 1Hosted: Cordon runs Cordon Cloud for you; nothing to operate.
- 2Your own cloud: the same platform inside your own cloud account, on any provider.
- 3Air-gapped: everything on your side, updated with signed bundles you carry in.