OWASP coverage

Where IntentGate sits in your OWASP defense

Two OWASP frameworks now apply to LLM and agent deployments: the Top 10 for LLM Applications, and the Top 10 for Agentic AI Applications. Below is honest, per-risk mapping of where IntentGate is a direct mitigation, a partial mitigation, or out of scope — with the specific mechanism in each case. For risks marked ★ Live demo, you can watch IntentGate block a real attack in our hosted lab.

See it blocked live → request lab access

Hosted lab. Real gateway. Fourteen scripted attacks across both OWASP frameworks. You drive; we provide the credentials.

14 Direct primary control via a specific IntentGate mechanism
1 Partial contributing evidence or scope, not the full answer
5 Out of scope needs other layers (model-time, supply-chain, code review)
01

OWASP Top 10 for LLM Applications

The original framework, focused on risks specific to language-model-backed applications. IntentGate is the runtime layer for LLM01, LLM02, LLM05, LLM06, LLM08, and LLM10 — six of the ten.

LLM01 Direct ★ Live demo

Prompt Injection

Intent check. The extractor surfaces divergence between the user's original ask and the agent's tool-call arguments. Injected instructions redirect the agent's action; this check catches the redirect.

Business impact if undefended: Attacker hijacks agent goal mid-flow; e.g. "customer service" agent is redirected to drop the customers table. GDPR Article 5 integrity breach; SOC 2 CC7.2.
LLM02 Direct ★ Live demo

Sensitive Information Disclosure

Bidirectional PII filter. Tool-call arguments scanned on the way out, tool-call responses scanned on the way back — same engine, same audit chain. Built-in classes (email, phone, IBAN, BSN, credit card, US SSN, EU VAT, IPv4/6) plus customer-declared patterns; each with a numeric validator (mod-97, mod-11, Luhn, SSA prefix rules) to keep false positives low. Three actions: redact, block, or escalate. Per-request override via Rego; counts-only in the audit chain — matched values are never persisted.

Business impact if undefended: Agent exfiltrates customer PII at scale. GDPR fine up to €20M or 4% of global revenue; mandatory 72-hour breach notification; brand impact.
LLM03 Out of scope

Supply Chain

Out of scope. IntentGate ships SBOMs and signed releases for its own components but doesn't address the customer's model and library supply chain.

LLM04 Out of scope

Data and Model Poisoning

Out of scope. Risk sits at training time; IntentGate is runtime.

LLM05 Direct ★ Live demo

Improper Output Handling

Output schema validation. Operators declare a JSON-Schema-subset shape per tool. Every upstream response is checked: undeclared fields stripped (default), wrong-type scalars dropped, enum violations refused. Three actions — allow, strip, block — with per-tool overrides. Runs after the PII filter on the response path, so PII-redacted text is also schema-checked before reaching the agent. Counts-only in audit; matched values never persisted.

Business impact if undefended: A tool spec'd to return {customer_id, status} silently leaks password_hash, ssn, or internal IDs because the agent has no way to enforce the contract. Cascades into LLM02 disclosure and downstream tool misuse. SOC 2 CC7.2; GDPR data-minimisation breach.
LLM06 Direct ★ Live demo

Excessive Agency

Capability tokens enforce least-privilege tool access. Rego policy enforces scope limits. Escalate decision gates high-impact actions for human approval.

Business impact if undefended: AI agent autonomously wires €99,999 with no human in the loop. EU AI Act Article 14 (human oversight) violation; banking license review; uninsurable under most cyber policies.
LLM07 Out of scope

System Prompt Leakage

Out of scope. Model-internal concern; IntentGate sits outside the model.

LLM08 Direct ★ Live demo

Vector & Embedding Weaknesses

Per-tenant scope enforcement. Operators mark vector-store and RAG tools as tenant-scoped. The gateway verifies the tool-call's tenant filter argument matches the verified capability token's tenant claim before forwarding. Missing filters can be auto-injected from the token; mismatched or wildcard filters are blocked (-32017). Closes the cross-tenant query path on shared embedding backends.

Business impact if undefended: An agent under tenant A queries the shared vector store with a wildcard filter and receives tenant B's confidential embeddings — the LLM02 disclosure pattern, one layer further back. GDPR Article 32 (security of processing); ISO 27001 A.13.1.3 (segregation in networks).
LLM09 Partial

Misinformation

Escalate decision routes high-stakes outputs to a human approver before they execute. Doesn't validate model factuality.

LLM10 Direct ★ Live demo

Unbounded Consumption

Budget check. Per-token call/cost ceilings, per-tenant quotas, circuit-breaker behavior on exhaustion.

Business impact if undefended: Runaway agent burns your cloud + LLM bill overnight. Documented industry incidents in the $30k–$100k range; no contractual recourse against the LLM provider.
02

OWASP Top 10 for Agentic AI Applications

The newer framework, specific to agents that plan, call tools, and act in multi-step workflows. IntentGate is the operational substrate this framework presumes — eight of the ten risks map to direct IntentGate mechanisms.

AGENT01 Direct ★ Live demo

Agent Goal Hijack

Intent check captures the user's original ask and compares to actual tool calls. Goal-drift from injected upstream data blocks at the gate.

Business impact if undefended: Agent silently changes goal mid-execution. Damage scales with the tools it has access to — billing system, customer DB, code-deploy pipeline. Reportable insider-pattern breach.
AGENT02 Direct ★ Live demo

Tool Misuse & Exploitation

Capability tokens allowlist tools per agent. Escalate routes sensitive tool calls to human approval with TOTP step-up. Audit chain monitors usage.

Business impact if undefended: Customer record deleted with no rollback. GDPR Article 5(1)(d) — integrity — reportable within 72 hours.
AGENT03 Direct ★ Live demo

Identity & Privilege Abuse

Short-lived capability tokens, per-agent identity, RBAC in the Pro console, continuous identity monitoring via audit chain.

Business impact if undefended: Read-only user promotes themselves to admin and mints unbounded tokens. ISO 27001 A.9.2.3 failure; full system takeover.
AGENT04 Out of scope

Agentic Supply Chain

Out of scope. Customer's responsibility to vet agent code and dependencies.

AGENT05 Out of scope

Code Execution (RCE)

Out of scope. IntentGate does not execute agent or model code.

AGENT06 Direct ★ Live demo

Memory & Context Poisoning

Opt-in memory provenance (enable per tenant): each memory entry the agent writes is signed with HMAC-SHA256 keyed by a per-session key derived (HKDF) from the capability token. At tool-call time the gateway re-derives the session key, verifies the HMAC over each entry the agent declares, and walks the per-session hash chain. Tampered entries — including the sophisticated case where the attacker swaps argument values but keeps the signature — fail verification before the call reaches the toolserver.

Business impact if undefended: AI agent wires legitimate-looking amounts to attacker-controlled accounts because the memory backend was tampered with. Wire-fraud reportable under EU 6th AMLD; restitution obligations to defrauded customers; SOX 404 ICFR finding if the agent has accounting access.
AGENT07 Direct ★ Live demo

Insecure Inter-Agent Comms

Capability-token attenuation provides byte-compatible signing and intent across agents. Gateway forces TLS. Audit chain captures every inter-agent call with root and parent JTIs.

Business impact if undefended: Forged inter-agent call grants attacker the same scope as the legitimate caller. IDOR-scale breach; reportable; full forensic investigation.
AGENT08 Direct ★ Live demo

Cascading Failures

Per-tool fault isolation. Bulkhead semaphores cap concurrent forwards per tool so one slow tool can't starve the rest. Per-tool circuit breaker opens after N consecutive upstream failures; fails-fast for the cooldown window, then probes via half-open. Independent state per tool — a breaker open on the billing tool doesn't affect the search tool. Combined with the budget check's per-token ceilings, this contains blast radius to a single failing dependency.

Business impact if undefended: One degraded tool (a vector store returning 500s, a database that's locking up) ties up every gateway goroutine and connection — every healthy tool inherits the slow tool's latency and the agent layer degrades together. Most outages CISOs care about start exactly this way. SOC 2 A1.2 (operational availability).
AGENT09 Direct

Human-Agent Trust Exploitation

Step-up authentication for sensitive actions. Approval workflows with TOTP/WebAuthn. Clear agent identity in audit chain so operators distinguish agent from human action.

AGENT10 Direct ★ Live demo

Rogue Agents

Capability tokens ARE the registration. Revocation list deny-lists rogue agents at the gate. Audit chain is the continuous inventory. Per-agent anomaly detection via SIEM queries.

Business impact if undefended: Ex-employee or compromised credential keeps working post-revocation. SOC 2 CC6.2 finding; pattern behind Capital One 2019 (100M records).

IntentGate is one layer in your defense-in-depth

Not a replacement for your AI security program. The runtime authorization layer that answers, for every tool call: "given everything we know about the originating request and the agent's identity right now — should this be allowed?"

Build time
Model and supply chain

SBOM hygiene, signed releases, vendor due diligence, model provenance, training-data validation.

LLM03 · LLM04 · LLM07 · AGENT04
Runtime
IntentGate

Capability, intent, policy, budget. Tamper-evident audit. Step-up for high-risk actions. The five-check authorization pipeline between every agent call and every tool.

14 direct · 1 partial mitigation
Agent runtime
Code and sandboxing

Sandbox tool execution, code review of agent orchestration, RCE prevention, dependency scanning.

AGENT05

A CISO evaluating IntentGate isn't replacing their entire AI-security program. They're filling in the layer that handles runtime authorization, intent verification, and tamper-evident audit — usually the layer that's most conspicuously missing today.