Policy engine
Rules by package name, version range, age, source, license and custom criteria. Block or log-only, with time-limited exceptions. Re-evaluated on every request.
The next generation Dependency Firewall is launching soon. See what’s new
Conda Dependency Firewall: Beta
Conda Dependency Firewall
Conda installs prebuilt binaries, resolved against a prioritized channel list. Whichever channel answers first wins, whether or not you meant it to. Decide what gets through.
Bytesafe Dependency Firewall sits in front of your existing repository, protecting developers, CI/CD pipelines and AI agents.
EU-based company · Software supply chain security since 2018.
bytesafe.dev / Firewall
47
Blocked today
1,284
Requests today
12
Active rules
Package
Rule
Status
malicious-pkg@2.1.0
Malware scan
BLOCKED
lodash@4.17.20
CVSS ≥ 7.0
BLOCKED
react@19.1.1
Allowlist
APPROVED
new-release@0.0.1
Age < 7 days
DELAYED
axios@1.7.9
Allowlist
APPROVED
Blocked
malicious-pkg@2.1.0
By Exception
lodash@4.17.21
Delayed
new-release@0.0.1
Approved
react@19.1.1
Supply chain attacks on Conda
A selection of documented supply chain attacks on Conda. The Conda ecosystem has been targeted by malicious packages, account takeovers, and dependency confusion attacks.
conda-forge token exposure2025
Misconfigured build infrastructure leaked the anaconda.org upload token for the conda-forge channel to every feedstock maintainer for about seven weeks (CVE-2025-31484). Anyone holding it could have uploaded or overwritten packages in the production channel. An audit found no packages were altered before the token was revoked.
Infrastructure exposureconda-forge-metadata2025
The conda-forge-metadata package declared an optional dependency, conda-oci-mirror, that had never been published under that name. Anyone could have claimed it and shipped code that ran on install, a flaw rated 9.3 CVSS before it was fixed.
Unclaimed dependencyxz backdoor check2024
When the CVE-2024-3094 backdoor in xz surfaced, conda-forge had to confirm its own builds were clean. Its last xz upload predated the compromised 5.6.x release line, so conda-forge artifacts were unaffected, but every team running a conda-distributed xz needed to run the same check.
Build toolchain backdoorDependency confusion2021+
Conda resolves package names against a prioritized list of channels. If a client's channel order isn't locked down, a public conda-forge or user-channel package can resolve ahead of an internal one with the same name.
Namespace attackDependency Firewall does not prevent all attacks, but blocks packages that match known malware signatures or violate your policies.
Intercepts every Conda package request before it reaches you.
Every package install is a potential entry point. Traditional SCA tools find problems after packages are already in your environment. Dependency Firewall intercepts every Conda request before it reaches your developers, CI/CD pipelines or AI agents.
You define the rules: block packages with known CVEs, block known malicious packages, or delay newly published versions for a configurable period to give the ecosystem community time to surface zero-day threats.
Works in front of enterprise repository platforms and any Conda registry. No agent installs. No workflow changes.
Public Conda registry
Vulnerable and malicious versions included
Risky packages
Dependency Firewall
Policy engineVetted packages only
Developers and CI/CD
Internal environment
Dependency Firewall capabilities
Policy controls, malware blocking, package delay, dependency confusion protection, publish scanning, and full audit visibility across every Conda request.
Vulnerability blocking
Block packages with known CVEs before install. Filter by CVSS and EPSS severity per registry or team. New advisories take effect immediately.
Malware scanning
Detect malicious payloads, suspicious install hooks and obfuscated code before execution. Quarantined packages are logged and never silently dropped.
Trust downgrade detection
Flags npm releases published with weaker provenance than the version before them, the pattern behind account-takeover attacks. npm only today.
Dependency confusion
Block namespace attacks where public packages impersonate your internal ones. Configurable upstream priority rules ensure private packages always win.
Zero-day safety delay
Hold newly published versions for a configurable window (7 or 14 days) before they reach developers or pipelines. Gives the ecosystem time to surface threats.
Package observations
Every package is fingerprinted: first-seen date, download frequency, requester, version age. Know exactly what passed through and when.
Audit logging
Every block, allow and exception is recorded and exportable to your SIEM. Built to make security teams and auditors happy out of the box.
Publish scanning
Packages are scanned for malware, secrets, and sensitive data before they are published to an upstream registry.
Beta limits
No vulnerability blocking yet
CVE and EPSS based rules aren't available for Conda. The dashboard doesn't offer them for Conda firewalls, and the API rejects a vulnerabilities rule.
License rules see unknown only
There's no license data source for Conda packages yet, so every version resolves as unknown license. Only the unknown option in a license rule has any effect.
Read-only channel
The firewall serves Conda channels read-only. There's no publish-scanning phase for uploads in the beta.
Add a security layer to Conda
Route Conda package traffic through Dependency Firewall, define policies for what's allowed and let the firewall block the rest. Developers and pipelines keep their existing package manager commands.
01
Route Conda requests through Dependency Firewall
Point your Conda configuration at Dependency Firewall. Every install request passes through the firewall before reaching the registry or your environment.
02
Define your security policies
Set vulnerability thresholds, enable malware scanning, configure safety delays for new versions and write allowlist or blocklist rules. Create multiple firewalls with individual rules for different teams or projects.
03
Bad packages are blocked. Safe ones flow through.
Every request is evaluated in real time. Blocked packages are logged with the policy that triggered them. Approved packages are served transparently.
Point conda, mamba or micromamba at Bytesafe Dependency Firewall
Set the firewall as the only channel with --override-channels, so no other configured channel or cached conda-forge entry can resolve packages around it. Credentials ride in the channel URL: HTTP basic auth with username __token__ and your Bytesafe access token as the password.
Works with the repositories you already use
JFrog Artifactory
Sonatype Nexus
GitLab
GitHub Packages
Azure Artifacts
AWS CodeArtifact
Firewall rules
Each rule targets an ecosystem and applies a condition: vulnerability severity, package age, license type or name pattern. Rules either block or log. Stack multiple rules per firewall. Changes take effect immediately.
Security teams can start with broad guardrails, then narrow policies by upstream, package, version range, internal status, maximum age, CVSS score and EPSS score.
Live firewall logs
Every blocked package is logged: package name, version, status, ecosystem, which firewall evaluated it, which rule triggered and who requested it. Filter by firewall or user.
The log view gives developers a fast answer when an install fails and gives AppSec a complete audit trail for policy enforcement.
Package details and scorecards
Open a package to review advisories, licenses, project metadata, OpenSSF Scorecard checks, dependency counts and source links before deciding whether to block, allow or investigate further.
The package page brings runtime firewall context together with upstream project health. Security teams can compare CVEs, maintainer signals, and repository hygiene from the same screen.
Sits in front of what you already run
Other enterprise dependency firewalls are often bundled into repository platforms. Dependency Firewall is an independent firewall that works with any registry and is built in the EU.
| Criterion | Dependency Firewall | Other enterprise firewalls |
|---|---|---|
| Works with your existing repository | Yes, as a proxy in front of it | Bundled into their platform most often |
| Deploys in minutes | Yes | Usually weeks of platform work most often |
| Predictable pricing | Flat, no usage-based fees | Usage-based or opaque most often |
| EU data sovereignty | Yes | No, US-based most often |
Conda Dependency Firewall: frequently asked questions
Common questions from security and engineering teams.
How do I point conda or mamba at Bytesafe Dependency Firewall?
Set the firewall as your only channel with --override-channels, or add it to channels in .condarc. Credentials go in the URL: username __token__, password your Bytesafe access token.
Does it work with mamba and micromamba?
Yes. mamba and micromamba read the same channel configuration as conda and are configured the same way.
What's different about Conda support in the beta?
Vulnerability blocking and license rules are limited: there's no vulnerability data source wired up yet, and license rules resolve every package as unknown. Publish scanning isn't available either, channels are read-only through the firewall for now.
Can I delay new Conda package versions the same way as npm or PyPI?
Yes. Age-based rules use the package's repodata timestamp, so you can hold newly published versions for a configurable window before they reach developers.
Can different projects have different policies?
Yes. You can create separate firewalls per project. They are lightweight and easy to clone. You can also differentiate by the user or token used for the session. Firewall configurations are small JSON files that can be managed in Git.
Can packages be delayed before they reach developers?
Yes. Dependency Firewall can hold newly published package versions for a configurable window (7 or 14 days) before they reach developers or pipelines. Centralizing delay rules means the protection applies automatically across all teams and pipelines without each project configuring it separately.
What happens if a package passes through but malware is found later?
The firewall tracks all packages via observations: first-seen date, last-seen date, and which firewalls they passed through. When new malware data surfaces, you can see exactly which projects downloaded the affected package and when.
Can firewall rules be automated?
Yes. All configuration is available via API. Configurations can be version-controlled in Git and deployed through your existing automation. All changes are tracked with full rollback support.
Does Dependency Firewall work with enterprise repository platforms?
Yes. Dependency Firewall speaks the same protocols as your package managers, so it is fully transparent to enterprise repository platforms and package registries, including JFrog Artifactory, Sonatype Nexus, GitLab, GitHub Packages and Azure Artifacts.
How is licensing structured?
Two plans: Cloud for SaaS and Enterprise for custom deployment, Managed Cloud or On-Premise. Both include unlimited users, package requests and bandwidth with no usage-based fees. See pricing for plan details and add-ons.
Watch it block Conda threats
Book a 30-minute session and we'll show you how Dependency Firewall fits into your existing setup. Your registries stay unchanged.