Rebuff

Open-source prompt-injection detector combining heuristics, an LLM check, a vector store and canary tokens.

Provider: default
Kind: Policy engine
Category: Operations
Official docs: https://github.com/protectai/rebuff

Properties · 3 set by this component

property	type	values	required	kind default	this component
canaryTokens · component property	boolean	—	no	false	inherited
effect	enum	audit \| deny \| modify \| deploy-if-not-exists	no	—	▸ audit
enforcementMode	enum	enforced \| disabled	no	—	inherited
heuristicCheck · component property	boolean	—	no	true	inherited
llmCheck · component property	boolean	—	no	true	inherited
managed	boolean	—	no	false	▸ false
policyCount	integer	—	no	—	inherited
remediation	boolean	—	no	false	inherited
scope	enum	resource \| group \| account \| organization	no	—	▸ resource
vectorCheck · component property	boolean	—	no	true	inherited

Rules that apply · 0

Every rule whose selector covers this component's kind or its category. Whether a rule fires depends on the design: its properties, its edges and its groups.

No rule reads this kind yet. That is a gap, not a guarantee.

Use it in a design

nodes:
  - { id: "rebuff", component: "rebuff" }

Other Policy engine products · 56

Cloud Configalibaba-cloud-configalibabacloudpolicy-engine Data Security Centeralibaba-data-security-centeralibabacloudpolicy-engine Amazon CodeGuru Revieweraws-codeguru-reviewerawspolicy-engine AWS Configaws-configawspolicy-engine AWS Control Toweraws-control-towerawspolicy-engine IAM Access Analyzeraws-iam-access-analyzerawspolicy-engine AWS Lake Formationaws-lake-formationawspolicy-engine AWS License Manageraws-license-managerawspolicy-engine AWS Organizationsaws-organizationsawspolicy-engine AWS Resource Access Manageraws-ramawspolicy-engine Amazon Verified Permissionsaws-verified-permissionsawspolicy-engine Microsoft Entra Conditional Accessazure-conditional-accessazurepolicy-engine Azure Policyazure-policyazurepolicy-engine Microsoft Entra Privileged Identity Managementazure-privileged-identity-managementazurepolicy-engine Cloudflare Data Loss Preventioncloudflare-dlpcloudflarepolicy-engine CalypsoAIcalypsoaipolicy-engine Cerboscerbospolicy-engine OPA Gatekeepergatekeeperpolicy-engine Guardrails AIguardrails-aipolicy-engine HiddenLayer AI Security Platformhiddenlayerpolicy-engine KubeArmorkubearmorpolicy-engine Kubewardenkubewardenpolicy-engine Kyvernokyvernopolicy-engine Lakera Guardlakera-guardpolicy-engine LLM Guardllm-guardpolicy-engine NeMo Guardrailsnemo-guardrailspolicy-engine Open Policy Agentopapolicy-engine Permit.iopermit-iopolicy-engine Generic policy enginepolicy-enginepolicy-engine Prompt Securityprompt-securitypolicy-engine Protect AIprotect-aipolicy-engine Robust Intelligencerobust-intelligencepolicy-engine SonarQube Serversonarqubepolicy-engine DigitalOcean CSPMdigitalocean-cspmdigitaloceanpolicy-engine Access Context Managergoogle-access-context-managergcppolicy-engine Assured Workloadsgoogle-assured-workloadsgcppolicy-engine Binary Authorizationgoogle-binary-authorizationgcppolicy-engine Identity and Access Managementgoogle-cloud-iamgcppolicy-engine Firebase App Checkgoogle-firebase-app-checkgcppolicy-engine Model Armorgoogle-model-armorgcppolicy-engine

and 16 more → Policy engine page

Source

1# yaml-language-server: $schema=../../schema/component.schema.json
2# Rebuff: Protect AI's open-source (Apache-2.0) prompt-injection detector layering heuristics, an LLM
3# classifier, a vector store of past attacks and canary tokens. Retired May 2025: the repo was archived
4# read-only on 16 May 2025 and called itself a prototype. Per python-sdk sdk.py, detect_injection runs all
5# three checks by default with thresholds 0.75 heuristic, 0.90 vector and 0.90 model.
6schemaVersion: "1"
7type: "component"
8id: "rebuff"
9kind: "policy-engine"
10provider: "default"
11docs: "https://github.com/protectai/rebuff"
12defaults:
13  managed: false
14  scope: "resource"
15  effect: "audit"
16properties:
17  heuristicCheck:
18    type: "boolean"
19    default: true
20  vectorCheck:
21    type: "boolean"
22    default: true
23  llmCheck:
24    type: "boolean"
25    default: true
26  canaryTokens:
27    type: "boolean"
28    default: false

The same page as markdown, for an agent: /catalogue/components/rebuff.md