Roundups of open-source tools for testing LLM applications and guarding them at runtime, judged on license, maintainer, coverage and fit with CI and production systems.
garak, PyRIT and promptfoo compared for probing LLM applications before release: attack coverage, scripting and CI integration, reporting, licenses and maintainers, and which combination fits a security team versus a product team.
NeMo Guardrails, LLM Guard and Guardrails AI compared for runtime protection: approach, threats covered, latency and model dependencies, licenses and the effort to run them inside your own infrastructure.
The maintained open-source options for teams still running LLM Guard after its July 2026 archive: NeMo Guardrails, Guardrails AI and Meta's LlamaFirewall and Prompt Guard, compared scanner by scanner with DeepTeam's guardrails named as a lighter option.
ModelScan, Fickling and picklescan for catching malicious pickle and model files before they load: formats, detection method, licenses, maintenance and where each fits in a model intake pipeline, with Hugging Face's own Hub scanning as context.
Tell us what you're working on, whether it's a deployment, an audit, a security test or a cyber range. You'll speak with an engineer who can help you scope it.