Threat Framework Coverage
Built against the frameworks your security team already uses.
OWASP, MITRE ATLAS, and MITRE ATT&CK define the threat landscape for AI systems. Every AInforce module maps directly to these controls — so your security team has a coverage story they can defend.
28 of 32 threats addressed across all three frameworks. Partial coverage noted where detection is present but automated response is in development.
Prompt Injection
Hundreds of injection variants tested per endpoint daily across direct and indirect attack surfaces. Auto-generated attack chains, scored and triaged.
Sensitive Information Disclosure
Every AI interaction logged with automated data classification. PII, source code, secrets, and regulated data flagged in real time with identity attribution.
Supply Chain
Model SBOM generation, cryptographic provenance verification, and dependency auditing across your entire AI stack — from base models to fine-tuned derivatives.
Data and Model Poisoning
Vector DB poisoning alerts, behavioral baseline drift detection, and hash-verified model integrity checks catch tampering before it reaches production inference.
Improper Output Handling
All LLM outputs inspected and classified before downstream consumption. Anomalous or policy-violating outputs flagged with full session context for review.
Excessive Agency
Tool call interception validates every action against a declared permission scope. Privilege escalation attempts blocked; human-in-the-loop approval queue for high-risk actions.
System Prompt Leakage
Daily test suites probe every AI endpoint for system prompt extraction via direct elicitation, role-play bypass, and jailbreak variants.
Vector and Embedding Weaknesses
Vector DB poisoning audits detect adversarial embeddings and data integrity violations in RAG pipelines. Baseline comparison surfaces drift over time.
Misinformation
Behavioural anomaly detection surfaces unusual output patterns. Ground-truth factual verification is on the roadmap for domain-specific knowledge bases.
Unbounded Consumption
Red Teaming runs token exhaustion and DoS variant suites daily. Agent Security enforces per-agent resource budgets and blocks runaway tool call chains.
Reconnaissance
Passive AI footprint mapping via DNS, certificate transparency, CSP headers, and ATS job board intelligence reveals your attack surface before adversaries do.
Resource Development
Attack infrastructure simulation covers most tooling categories. Tracking of adversarial model development (shadow model training) is on the roadmap.
Initial Access
Scanner identifies externally exposed AI endpoints. Agent Security enforces access controls at the runtime layer. Combined coverage of both external and internal access paths.
Execution
OWASP LLM attack suites simulate adversarial execution paths daily — including prompt-injected code execution, tool call abuse, and agentic chain exploitation.
Persistence
Cryptographic model integrity verification and behavioral baseline comparison detect unauthorized model modifications that would enable persistent attacker access.
Privilege Escalation
Real-time privilege escalation detection blocks agents from acquiring permissions beyond their declared scope. Multi-step chain auditing catches escalation sequences.
Defense Evasion
Jailbreak, role-play bypass, and context-window manipulation test suites simulate evasion techniques. Continuously updated attack library tracks new evasion research.
Discovery
External reconnaissance from the attacker's vantage point — enumerating your AI tools, inference endpoints, and SaaS integrations before an attacker does.
Collection
Complete AI interaction logging with PII, code, and secrets classification. Every data element moving through an AI system is captured, labelled, and attributed to an identity.
ML Attack Staging
Daily OWASP attack suite runs simulate full attack staging sequences — reconnaissance to exploit — against each AI endpoint in your environment.
Exfiltration
Output inspection classifies every model response for sensitive data. Anomalous exfiltration patterns trigger alerts with the full session and identity context.
Impact
High-risk agentic actions require human-in-the-loop approval before execution. Tool call blocking and rate enforcement stop downstream impact before it occurs.
Exploit Public-Facing Application
AI endpoints treated as public-facing attack surfaces. Daily exploit suites test LLM APIs, agent endpoints, and RAG interfaces for injection, extraction, and bypass vulnerabilities.
Command and Scripting Interpreter (via AI agents)
Tool call validation intercepts every code execution request issued by an AI agent. Declared-scope enforcement blocks unapproved interpreter access before execution.
Supply Chain Compromise
Model SBOM tracking, cryptographic provenance chains, and hash-verified integrity checks cover the AI model supply chain from pre-trained base to fine-tuned production version.
Unsecured Credentials (in AI prompts and outputs)
Real-time secrets detection in every AI interaction. Credentials, API keys, and tokens embedded in prompts or returned in model outputs are flagged immediately.
Data from Information Repositories (RAG exfiltration)
Supply Chain audits RAG pipeline integrity. Observatory classifies data moving through retrieval-augmented queries and detects over-permissive retrieval patterns.
Data Manipulation (model and vector poisoning)
Vector DB poisoning audit and model behavioral drift detection identify adversarial data manipulation before it affects production inference or downstream decisions.
Exfiltration Over Alternative Protocol (via AI APIs)
AI API interactions logged and inspected for anomalous data volumes and sensitive content. Exfiltration via model outputs detected regardless of the API transport.
Endpoint Denial of Service (via AI agents)
Agent Security enforces per-agent compute and call budgets. Red Teaming runs token exhaustion and recursive tool call DoS simulations against every AI endpoint.
Valid Accounts (AI-assisted credential theft)
Behavioural anomaly detection flags account usage patterns inconsistent with the established baseline. Full identity provider integration for anomalous login correlation is on the roadmap.
Phishing (AI-enhanced spearphishing)
Employee AI tool usage patterns surfaced for policy enforcement. Detection of AI-generated phishing content in outbound interactions is on the roadmap.
Coverage legend
Coverage is self-attested and reflects the capabilities of the Monitor and Enforce tiers. The Free tier (AI Risk Scanner) covers OWASP LLM Top 10 external surface indicators only. ATT&CK technique selection focuses on AI-specific abuse paths — the full ATT&CK matrix contains 200+ techniques; contact us for a complete mapping.
NIST AI RMF, ISO 27001, SOC 2, GDPR, EU AI Act