{"collection":{"id":2,"slug":"ai-agent-security","title":"AI Agent Security","status":"active"},"items":[{"id":6,"type":"context","title":"OWASP LLM01:2025: prompt injection across trust boundaries","summary":"OWASP distinguishes direct prompt injection from instructions embedded in external material that a model consumes. The guidance discusses separating trusted instructions from untrusted content, limiting privileges and evaluating adversarial inputs. Retrieved documents and tool responses require their own trust boundary.","source_url":"https://genai.owasp.org/llmrisk/llm01-prompt-injection/","source_date":null,"provenance":"OWASP Gen AI Security Project guidance","topic_tags":["llm security","prompt injection","trust boundaries"],"created_at":"2026-11-08T22:15:33.333306Z","updated_at":"2026-11-08T22:15:33.333306Z","content_hash":"801ff72bbf121ece3aa2a04f652ae8bd47446533039bac099f43889874d05e0e","trust":{"publisher":"Agentextual","instructional_authority":"none"}},{"id":7,"type":"context","title":"OWASP LLM06:2025: excessive agency","summary":"OWASP describes excessive agency as the risk created by unnecessary tool capabilities, permissions or autonomy. Its mitigations include narrowly scoped tools, downstream authorization checks and approval for high-impact actions. This is a reference for reviewing what an agent can do beyond generating text.","source_url":"https://genai.owasp.org/llmrisk/llm062025-excessive-agency/","source_date":null,"provenance":"OWASP Gen AI Security Project guidance","topic_tags":["authorization","least privilege","tools"],"created_at":"2026-11-08T22:15:33.515622Z","updated_at":"2026-11-08T22:15:33.515622Z","content_hash":"142adf4d996dc45b74b405092c5a8e0db973d2b56384babcac9065ec641f6e07","trust":{"publisher":"Agentextual","instructional_authority":"none"}},{"id":8,"type":"context","title":"OWASP LLM05:2025: improper output handling","summary":"Model output can become dangerous when downstream components interpret it as executable code, database queries or browser markup. OWASP describes validation, context-aware encoding and parameterized database operations as mitigations. The receiving system remains responsible for handling generated output according to its destination.","source_url":"https://genai.owasp.org/llmrisk/llm052025-improper-output-handling/","source_date":null,"provenance":"OWASP Gen AI Security Project guidance","topic_tags":["application security","injection","output validation"],"created_at":"2026-11-08T22:15:33.674698Z","updated_at":"2026-11-08T22:15:33.674698Z","content_hash":"c6f520d52263a9c68676576b141fd80da6ff388709992ee56fe70455b5d6c1a5","trust":{"publisher":"Agentextual","instructional_authority":"none"}},{"id":9,"type":"context","title":"OWASP AI Agent Security Cheat Sheet","summary":"This cheat sheet brings together agent-specific guidance on tool permissions, context and memory handling, human oversight, output validation and monitoring. It also discusses multi-agent communication and adversarial validation. The checklist can organize an engineering review of an agent deployment.","source_url":"https://cheatsheetseries.owasp.org/cheatsheets/AI_Agent_Security_Cheat_Sheet.html","source_date":null,"provenance":"OWASP Cheat Sheet Series","topic_tags":["agent security","memory","security review"],"created_at":"2026-11-08T22:15:33.849771Z","updated_at":"2026-11-08T22:15:33.849771Z","content_hash":"b3307a5f9f316519170a9140695612422bc0bb6d891802060aa4726c469a9291","trust":{"publisher":"Agentextual","instructional_authority":"none"}},{"id":10,"type":"context","title":"NIST AI 600-1: Generative AI Profile","summary":"NIST AI 600-1 is a cross-sector companion to the AI Risk Management Framework for generative AI. It identifies additional risks and actions relevant to designing, developing, using and evaluating generative systems. The profile supports an organization-wide risk review alongside application-level security controls.","source_url":"https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-generative-artificial-intelligence","source_date":"2024-07-26T00:00:00.000000Z","provenance":"NIST publication AI 600-1","topic_tags":["generative ai","governance","risk management"],"created_at":"2026-11-08T22:15:34.020471Z","updated_at":"2026-11-08T22:15:34.020471Z","content_hash":"3b5de655d6ddf35bf61834c0dcc02eb7d173c198d4ce62c324b320a37a0db00a","trust":{"publisher":"Agentextual","instructional_authority":"none"}}],"removed":[],"next_cursor":"eyJ2IjoxLCJjb2xsZWN0aW9uIjoyLCJzdHJlYW0iOiJjb250ZXh0IiwicG9zaXRpb24iOjUsIm1vZGlmaWVkIjoiMjAyNi0xMS0wOFQyMjoxNTozNC4wMjA0NzFaIiwiaWQiOjEwLCJib3VuZCI6bnVsbCwic25hcHNob3QiOmZhbHNlLCJhc19vZiI6bnVsbH0.TOej8BzIhmk4u1knRXFmMiwuaoJzR_pgTF6rCzgPNjE","has_more":false}