OWASP Top 10 for LLM Applications 2026
LLM01:2026 Prompt Injection
0 advisories on ShieldMap fall under it through CWE-1427, and 3 technique pages explain how the attacks work.
OWASP Top 10 for Agentic Applications
- ●ASI01:2026 Agent Goal HijackInjected input overrides the system-prompt role/capability constraints, redirecting agent goals.
- ●ASI02:2026 Tool Misuse & ExploitationInjected input drives unauthorized tool invocation (the "lethal trifecta" conditions).
- ●ASI03:2026 Identity & Privilege AbuseAgent acts under the user's elevated credentials, performing actions the attacker could not.
- ●ASI05:2026 Unexpected Code Execution (RCE)Shell / file-system / cloud-API access turns injection into arbitrary command execution.
- ●ASI06:2026 Memory & Context PoisoningMemory and RAG poisoning taint every future session reading the store.
- ●ASI08:2026 Cascading FailuresTool outputs re-enter the context window, enabling chained or self-replicating effects.
- ●ASI09:2026 Human-Agent Trust ExploitationInjection bypasses human-in-the-loop confirmation.
CWE weaknesses
- ●CWE-1427 Improper Neutralization of Input Used for LLM PromptingUntrusted input from prompts, retrieved documents, tool output, or memory alters model behavior because the model draws no boundary between instructions and data, the definitional prompt-injection weakness.
- ○CWE-707 Improper NeutralizationThe pillar-level root cause is a failure to separate and neutralize attacker instructions from data on a single token stream.
- ○CWE-349 Acceptance of Extraneous Untrusted Data With Trusted DataContext-window pooling merges untrusted retrieved content and tool output with trusted instructions under no enforced trust boundary.
- ○CWE-693 Protection Mechanism FailureJailbreaks and adaptive guardrail evasion degrade prompt-injection classifiers and link filters, driving defenses to high adversarial success rates.
MITRE ATLAS tactics
- ●AML.TA0004 Initial AccessInjected instructions, whether typed directly into the prompt or hidden in content the model later retrieves, are the foothold that lets an attacker seize control of the model's behavior.
- ●AML.TA0005 ExecutionA successful injection drives the agent to invoke tools and run adversary-chosen commands, including arbitrary code execution on the host.
- ●AML.TA0006 PersistenceInjected instructions written into long-term memory or a retrieval corpus survive across sessions and re-fire whenever the poisoned entry is recalled.
- ●AML.TA0007 Defense EvasionAttackers hide the injection with invisible Unicode, Base64 or ROT13 encoding, and low-resource languages, and use jailbreak phrasing to slip past safety filters.
- ●AML.TA0010 ExfiltrationInjection redirects the model to leak private conversation, database, or repository contents through image-URL, tool, or invisible-character channels.
- ○AML.TA0001 AI Attack AdaptationAdversaries optimize the injection payload in advance, using gradient-oracle fine-tuning and payload-splitting to raise attack success rates.
- ○AML.TA0011 ImpactA runtime injection can trigger destructive action, such as wiping a developer's local files.
- ○AML.TA0012 Privilege EscalationAn injected agent running on a trusted backend acts under the user's elevated credentials, yielding access beyond the attacker's own rights.
NIST AI 600-1 risks
- ●NIST Information SecurityPrompt injection expands the attack surface, letting adversarial text hijack the model to exfiltrate data or trigger unauthorized actions across connected systems.
- ○NIST Information IntegrityInjected instructions manipulate outputs into attacker-chosen content and poison retrieved or remembered context, corrupting the information the system produces.
- ○NIST Data PrivacySuccessful injection can exfiltrate private conversation history, uploaded documents, and repository contents to an attacker.
- ○NIST Human-AI ConfigurationHuman-in-the-loop confirmation degrades under approval fatigue, and benign content can carry unintended injected instructions, both matters of human-oversight configuration.
NIST AI RMF categories
- ○MEASURE 1 methods & metricsThe entry rejects static-only attack-success-rate claims and mandates adaptive-attack metrics, baselining injection defenses with AgentDojo and JailbreakBench because static scores run near zero while adaptive success exceeds 90 percent.
- ○MEASURE 2 evaluate trustworthinessDefenses are red-teamed against attackers who have read the full defense specification, evaluating the deployed system's injection resilience rather than trusting a metric selected in isolation.
- ○MAP 4 map risks across all componentsDecomposing an injection along delivery surface, propagation behavior, and encoding is presented as a threat-modeling step that maps risk across components including RAG corpora, MCP servers, and third-party tool packages.
OWASP GenAI Data Security
- ●DSGAI01 Sensitive Data LeakageA successful injection makes disclosure and exfiltration the dominant outcome, spilling system-prompt content, retrieved private documents, and infrastructure details out of the model.
- ●DSGAI06 Tool, Plugin & Agent Data Exchange RisksInjection rides through MCP servers and third-party tool packages, where poisoned tool descriptions and unpinned connections let a compromised model act across connected systems.
- ○DSGAI09 Multimodal Capture & Cross-Channel Data LeakageSteganographic image and invisible-Unicode payloads deliver instructions the human never sees, and markdown image-URL rendering carries the stolen data back out.
- ○DSGAI16 Endpoint & Browser Assistant OverreachBrowser, IDE, and email assistants that auto-summarize untrusted pages become the indirect-injection channel, from zero-click document exfiltration to a flipped IDE configuration flag.
- ○DSGAI04 Data, Model & Artifact PoisoningOne tainted entry in a RAG corpus or persistent memory taints every later session that reads it, making poisoning a cross-session propagation path for injection.
OWASP AIVSS agentic risks
- ●AIVSS-10 Agent Goal and Instruction ManipulationA crafted message overrides the system prompt's role and capability limits, redirecting the model to act outside its intended scope.
- ●AIVSS-1 Agentic AI Tool MisuseInjected input drives unauthorized invocation of tools the agent is permitted to call, from file system and shell to email and cloud APIs.
- ●AIVSS-2 Agent Access Control ViolationThe agent reads attacker-planted text while operating under the user's elevated credentials and performs privileged actions the attacker could not reach directly.
- ●AIVSS-6 Agent Memory and Context ManipulationAn injection written to long-term memory or a RAG corpus taints every later session that reads the poisoned store, persisting the compromise across conversations.
- ○AIVSS-7 Insecure Agent Critical Systems InteractionWhere the agent holds shell, file-system, or cloud-API access, an injection escalates into arbitrary command execution and destructive actions on the host.
- ○AIVSS-3 Agent Cascading FailuresTool outputs re-enter the shared context window, letting one injection chain into further tool calls or self-replicate across steps.
- ○AIVSS-4 Agent Orchestration and Multi-Agent ExploitationAn injection can self-replicate across agents, spreading from one compromised agent to its peers in a multi-agent workflow.
CSA AI Controls Matrix domains
- ●AIS Application & Interface Security
- ○IAM Identity & Access Management
- ○TVM Threat & Vulnerability Management
- ○STA Supply Chain Management, Transparency, and Accountability
Mappings and their reasons from the OWASP Top 10 for LLM Applications 2026, Appendix A (OWASP GenAI Security Project), shared under CC BY-SA 4.0. Mappings restructured as data; the ATT&CK section and the AICM relevance text left out. Navigation, not a compliance claim.