NIST AI RMF
The NIST AI Risk Management Framework 1.0 has four core functions: Govern, Map, Measure and Manage. A rule names the function its control relates to; only 95 of 1562 rules (rules 0.8.0) carry the mapping, because it applies to controls, not to every defect. It is a reference, not a conformance claim.
This is an edge map, not a score. A count says how many rules look at a class of weakness and how sure they are, not how safe a system is; a class with few rules may simply need few, and no number here means a framework is satisfied. Confidence: P0 exploitable as written, P1 dangerous in context, P2 heuristic or posture. Edges are the relationships the rules inspect (reaches, calls, reads, writes, sends, grants, depends_on, trusts).
Coverage matrix
Section titled “Coverage matrix”| Entry | Rules | P0 | P1 | P2 | Edges |
|---|---|---|---|---|---|
| GOVERN Govern | 50 | 0 | 24 | 26 | reaches 3 · calls 8 · reads 2 · writes 4 · sends 6 · grants 19 · depends_on 5 · trusts 1 · none 2 |
| MAP Map | 27 | 0 | 9 | 18 | calls 1 · writes 4 · grants 15 · depends_on 6 · none 1 |
| MEASURE Measure | 25 | 0 | 12 | 13 | calls 11 · reads 2 · writes 2 · sends 2 · grants 1 · depends_on 2 · trusts 1 · none 4 |
| MANAGE Manage | 86 | 0 | 30 | 56 | reaches 3 · calls 13 · reads 4 · writes 9 · sends 8 · grants 41 · depends_on 5 · trusts 1 · none 2 |
GOVERN Govern
Section titled “GOVERN Govern”50 rules · confidence P0 0, P1 24, P2 26 · severity error 0, warning 23, info 27
Edges inspected: reaches 3 · calls 8 · reads 2 · writes 4 · sends 6 · grants 19 · depends_on 5 · trusts 1 · none 2
30 of 50 rules, surest first
- seceng.agentframework.crewai-delegation-with-dangerous-tools
- seceng.cloudai-cross-account-model-invocation
- seceng.cloudai.vertex-broad-aiplatform-permissions
- seceng.cloudai.vertex-public-endpoint-or-logging
- seceng.langgraph-human-interrupt-bypass
- seceng.redis-url-without-tls
- seceng.cloudai.azure-openai-public-network-access
- seceng.cloudai.bedrock-prompt-logging-or-tracing
- seceng.dataset.eval-set-from-production-no-redaction
- seceng.dataset.hf-unknown-user-repo-for-finetune
- seceng.dataset.synthetic-data-leaks-source-records
- seceng.forensics-no-user-tenant-on-ai-event
- seceng.forensics-prompt-hash-missing
- seceng.forensics.ai-action-missing-correlation-id
- seceng.forensics.tool-call-missing-audit-event
- seceng.langgraph-state-persists-sensitive-context
- seceng.promptsc-runtime-prompt-edit-route
- seceng.promptsc.env-controlled-system-prompt
- seceng.promptsc.prompt-template-from-database-without-version
- seceng.promptsc.yaml-agent-config-remote-tools
- seceng.redaction-langsmith-langfuse-capture-raw-prompts
- seceng.redis-public-cache-no-auth
- seceng.toolauth.audit-missing-for-dangerous-tool
- seceng.vendor.openai-prompt-cache-sensitive-context
- seceng.agentframework.autogen-code-execution-enabled
- seceng.agentframework.autogen-human-input-never-with-tools
- seceng.cloudai.azure-openai-key-in-client-env
- seceng.cloudai.bedrock-broad-invoke-model-policy
- seceng.forensics-sensitive-ai-audit-raw
- seceng.promptsc.langchain-hub-prompt-unpinned
MAP Map
Section titled “MAP Map”27 rules · confidence P0 0, P1 9, P2 18 · severity error 0, warning 11, info 16
Edges inspected: calls 1 · writes 4 · grants 15 · depends_on 6 · none 1
27 rules
- seceng.agentframework.autogen-shared-workdir-or-history
- seceng.agentframework.crewai-memory-without-tenant-scope
- seceng.dataset.hf-unknown-user-repo-for-finetune
- seceng.dataset.remote-training-data-no-integrity
- seceng.forensics-model-version-missing
- seceng.langgraph.stategraph-without-typed-state
- seceng.promptsc.yaml-agent-config-remote-tools
- seceng.redis-embedding-cache-shared-key
- seceng.vectordb.shared-collection-name
- seceng.dataset.hf-dataset-unpinned
- seceng.promptsc.langchain-hub-prompt-unpinned
- seceng.vectordb.chroma-query-without-where
- seceng.vectordb.chroma-query-without-where-python
- seceng.vectordb.lancedb-search-without-filter
- seceng.vectordb.milvus-search-without-expr
- seceng.vectordb.milvus-search-without-expr-js
- seceng.vectordb.pinecone-query-without-namespace-or-filter
- seceng.vectordb.qdrant-query-without-filter
- seceng.vectordb.qdrant-query-without-filter-js
- seceng.vectordb.weaviate-query-without-where-filter
- seceng.langgraph-checkpoint-shared-or-unscoped
- seceng.promptsc-mcp-manifest-from-url
- seceng.promptsc.remote-prompt-url-runtime
- seceng.redis.ai-cache-without-tenant-prefix
- seceng.vendor.openai-assistants-thread-not-user-bound
- seceng.vendor.openai-file-search-vector-store-no-boundary
- seceng.vendor.openai-realtime-without-auth-or-rate-limit
MEASURE Measure
Section titled “MEASURE Measure”25 rules · confidence P0 0, P1 12, P2 13 · severity error 0, warning 10, info 15
Edges inspected: calls 11 · reads 2 · writes 2 · sends 2 · grants 1 · depends_on 2 · trusts 1 · none 4
25 rules
- seceng.schema-model-score-controls-security-decision
- seceng.dataset.eval-set-from-production-no-redaction
- seceng.dataset.remote-training-data-no-integrity
- seceng.dataset.synthetic-data-leaks-source-records
- seceng.forensics-ai-error-swallowed
- seceng.forensics-model-version-missing
- seceng.forensics-no-user-tenant-on-ai-event
- seceng.forensics-prompt-hash-missing
- seceng.forensics.ai-action-missing-correlation-id
- seceng.forensics.tool-call-missing-audit-event
- seceng.langgraph.stategraph-without-typed-state
- seceng.toolauth.audit-missing-for-dangerous-tool
- seceng.dataset.hf-dataset-unpinned
- seceng.redis-cache-poisoning-from-model-output
- seceng.schema-model-output-controls-action
- seceng.schema-model-output-controls-orm-filter
- seceng.schema-structured-output-non-strict
- seceng.schema.model-json-parse-without-validation-python
- seceng.schema.model-json-parse-without-validation-ts
- seceng.stream-html-markdown-raw-stream
- seceng.stream-partial-json-command-execution
- seceng.langgraph-conditional-edge-from-model-output
- seceng.schema-openai-json-mode-without-schema
- seceng.stream.sse-direct-llm-stream-no-filter-ts
- seceng.stream.websocket-direct-llm-output
MANAGE Manage
Section titled “MANAGE Manage”Some rules name a category inside the function (MANAGE-2.4); they are counted here and in the function.
86 rules · confidence P0 0, P1 30, P2 56 · severity error 2, warning 45, info 39
Edges inspected: reaches 3 · calls 13 · reads 4 · writes 9 · sends 8 · grants 41 · depends_on 5 · trusts 1 · none 2
30 of 86 rules, surest first
- keryn.failopen.py.gate-exception-runs-action
- keryn.failopen.ts.gate-exception-runs-action
- seceng.agentframework.crewai-delegation-with-dangerous-tools
- seceng.cloudai-cross-account-model-invocation
- seceng.cloudai.vertex-broad-aiplatform-permissions
- seceng.cloudai.vertex-public-endpoint-or-logging
- seceng.langgraph-human-interrupt-bypass
- seceng.redis-url-without-tls
- seceng.schema-model-score-controls-security-decision
- seceng.agentframework.autogen-shared-workdir-or-history
- seceng.agentframework.crewai-memory-without-tenant-scope
- seceng.agentframework.crewai-process-without-rate-or-step-limit
- seceng.cloudai.azure-openai-public-network-access
- seceng.cloudai.bedrock-prompt-logging-or-tracing
- seceng.dataset.remote-training-data-no-integrity
- seceng.forensics-ai-error-swallowed
- seceng.forensics.ai-action-missing-correlation-id
- seceng.forensics.tool-call-missing-audit-event
- seceng.langgraph-state-persists-sensitive-context
- seceng.langgraph.stategraph-without-typed-state
- seceng.loop.streaming-without-cancel-or-timeout
- seceng.promptsc-runtime-prompt-edit-route
- seceng.promptsc.env-controlled-system-prompt
- seceng.promptsc.prompt-template-from-database-without-version
- seceng.redaction-langsmith-langfuse-capture-raw-prompts
- seceng.redis-embedding-cache-shared-key
- seceng.redis-public-cache-no-auth
- seceng.toolauth.audit-missing-for-dangerous-tool
- seceng.vectordb.shared-collection-name
- seceng.vendor.openai-prompt-cache-sensitive-context