GEN AI SECURITY LANDSCAPE SUBMISSION Solutions Landscape Submission GEN AI SECURITY LANDSCAPE SUBMISSION Solutions Landscape Submission About Agentic GenAI LLM Red Teaming Submitter DetailsSubmitter details are used for any entry clarification required.Submitter eMailSubmitter AffiliationEnd UserCompanySolution DetailsTitle * Open Source Commercial Proprietary Solution Name * Company / Project NameSolution Description * Visual Text Solution LinkFeatured Image No Image selected Add Image TitleAlternative TextLeave empty if the image is purely decorative.CaptionDescriptionUpdate Image 0% Github DetailsGitHub StarsGitHub ForksCoverage DetailsLifecycle Stage (s) * Govern Monitor Operate Deploy Release Test & Evaluate Develop & Experiment Augm & Fine Tune Data Scope & Plan Test & Evaluation Augm & Fine Tune Data Agentic Top 10 Coverage * ASI02:26 ASI03:26 ASI05:26 ASI06:26 ASI07:26 ASI01:26 ASI08:26 ASI02:26 ASI09:26 ASI03:26 ASI10:26 ASI04:26 ASI05:26 ASI06:26 ASI07:26 ASI08:26 ASI09:26 ASI10:26 ASI01:26 Capabilities Coverage * Data Augmentation & Fine-Tuning Apply differential privacy or obfuscation on sensitive data injected into agent memory. Data masking on structured data. Agent Action Audit Detect usage of poisoned models in training pipelines Apply PII & Sensitive data masking injected into agent components Development & Experimentation Perform SAST/DAST on agent planning code, tool wrappers, & plugin interfaces. Harden agent loop logic against infinite loops, unsafe function routing, & unauthorized self-modification. Validate connector (e.g., MCP) contracts (input/output schemas & permissions). Implement policy enforcement hooks in Frameworks (e.g. LangGraph, CrewAI, Others) Test & Evaluation Agent Penetration Testing Adversarial red-teaming: goal drift, prompt injection, hallucination chaining, & over-permissioned tool usage. Multi-agent scenario simulations for collusion, misalignment, or deception detection. Validate agent decisions against expected goal plans. Sandboxed testing of all tool calls, code execution, cloud API triggers Available Agent Scanning Scoping/Planning Identify system-wide non-human Identities & Auth Protocols Draft policy for Agent privilege boundaries Draft policy for delegation logic Define controls for memory scoping, isolation & long-term persistance Support for Gen AI Security Project - Agentic Security Threat Modeling Approach Conducting Agentic Threat Modeling Draft policy Agent for tool scopes Scope & Plan Augm & Fine Tune Data Develop & Experiment Test & Evaluate Release Generate & verify model + agent + tool SBOMs - shared responsibility Register all agents in an internal trust registry Sign model weights, plugin manifests, & memory snapshots. Deploy Rotate all shared secrets, keys, & tokens with ephemeral, scoped credentials. Enforce zero-trust policies between agents, tools, & external APIs Configure Inter-agent authorization policies, capabilities, & roles Apply & manage runtime Guardrails Monitor Audit reflection accuracy by comparing stated & observed planning outcomes. Use immutable logs (e.g., Sigstore, Immudb) for forensic readiness. Alert on anomalies; e.g., goal reversal, unexpected plan depth, adversarial-input, excessive tool usage, or rapid inter-agent chatter. Correlate telemetry from agent step tracing, tool execution, & message logs. Operate Monitor agent memory mutation patterns for drift Detect task replay, infinite delegation, or hallucination loops. Enable human-in-the-loop (HITL) override thresholds on high-risk or ambiguous actions LLM Incident Detection & Response Continuously scan loaded plugins for CVEs & privilege escalation vectors. Runtime guardrails & moderation; anomalous tool use Monitor agent memory mutation patterns for drift, LLM Incident Detection & Response, Runtime guardrails & moderation; anomalous tool use Govern Automate agent versioning, expiration, & rotation policies. Enforce role- & task-based access policies across agent populations & their tool access. Align control evidence with frameworks like EU AI Act, NIST AI RMF, & ISO/IEC 42001. Automate goal alignment audits, including adversarial review of long-term agent memory. Taxonomy * reCaptcha Submit
