Security resources
Free AI and software supply chain security resources
Guides, templates and checklists from the PexLens Research Team. Use them to set AI policy, vet new tools and keep malicious packages, extensions and models out of your environment.
Guides and templates
Start here
AI security policy template
A free AI acceptable use and security policy you can adapt and copy today.
Get the template →Prompt injection protection
How prompt injection works, real attack paths and the controls that limit damage.
Read the guide →Agentic AI security
Securing AI agents and MCP servers with least privilege, approvals and monitoring.
Read the guide →PexLens FAQ
Answers about AI security, the PexLens products, pricing and deployment.
Browse the FAQ →Security advisories
Important package and ecosystem findings from the PexLens Research Team.
View advisories →Checklists
Quick checks before you say yes
Use these short checklists in tool reviews and change approvals. They reflect the attack patterns our researchers see most often.
Before you approve an AI tool
Ask the vendor, and check in the admin console:
- Is our data retained or used for training, and can that be turned off?
- Does it support single sign-on and admin controls?
- Where is data processed and stored?
- Can usage and audit logs be exported?
- Which connectors, plugins and agents can it use?
- Is there a data processing agreement and a security contact?
Before you install an open-source package
Check the package, not just the name:
- Name is not a near-miss of a popular package (typosquatting)
- Publisher, age and download history look normal
- No unexpected install scripts (postinstall, setup.py)
- No recent maintainer change or sudden major version jump
- No unexplained network, shell or credential access
- Known vulnerabilities and licence reviewed (SCA)
Before you load an AI model
Treat model files like executable code:
- Source and publisher on the model hub are verified
- Prefer safe formats such as safetensors over pickle
- File scanned for embedded code before loading
- Licence and intended use fit your case
- Exact revision pinned, not just the latest
- Recorded in your AI-BOM
Before you allow a browser or IDE extension
Look at what it can do, and what it might do later:
- Permissions requested match its purpose
- Publisher is verified and established
- Update history shows no sudden permission changes
- Network destinations are known and expected
- Who in the organisation already uses it
- Version pinned and updates reviewed
Glossary
AI and supply chain security terms, explained
- AI DLP
- AI data loss prevention: inspecting prompts, uploads and responses so sensitive data does not reach AI tools.
- SEB
- Secure enterprise browser: a managed browser that applies security policy to web and SaaS work.
- SSCS
- Software supply chain security: protecting the code, packages, tools and pipelines used to build software.
- SCA
- Software composition analysis: finding open-source components and their known vulnerabilities and licences.
- SBOM
- Software bill of materials: an inventory of the components inside a piece of software.
- AI-BOM
- AI bill of materials: an inventory of the models, datasets, prompts and AI services a system depends on.
- MCP
- Model Context Protocol: a standard way for AI agents to connect to tools and data sources.
- Prompt injection
- Instructions hidden in content an AI reads that make it ignore its rules or take unintended actions.
- Typosquatting
- Publishing a malicious package with a name close to a popular one, hoping for a mistyped install.
- Dependency confusion
- Tricking a build into installing a public package that shares a name with an internal one.
- Extension drift
- A trusted extension changing behaviour, permissions or network destinations after an update.
- Model serialization attack
- Hiding code in a model file format, such as pickle, so it runs when the model is loaded.
Automate the checks
PexLens applies these checks automatically to AI prompts, browser sessions, packages, extensions and models.