AI Search / RAG failure assessment

Search / RAG failure surface

• 66failures on this surface
• 4catastrophic
• 32%under active regulatory exposure

1. Hallucination

   36 on this surface
   1 Catastrophic 14 High 20 Medium 1 Low

   Runtime control Prism observes hallucination signatures in the model's internal state. AIDR flags the moment the model commits to a fabricated claim. OmniGuard can block the response inline.

   • Oregon attorneys fined $110,000 for AI-generated fake case lawCatastrophic Dec 2025
   • California judge relied on fictitious AI case law in H.C. v. ContrerasHigh May 2026
   • An attorney in Dubinin v. Papazian filed a brief with ten AI-fabricated citations, ending the caseHigh Nov 2025

2. Brand & Safety Incident

   14 on this surface
   1 Catastrophic 6 High 6 Medium 1 Low

   Runtime control Prism reads the model's representation against brand and safety policy. OmniGuard blocks inline. AIDR provides the post-incident audit trail.

   • Sports Illustrated published AI-generated articles under fake author namesCatastrophic Nov 2023
   • OpenAI's Sora app filled with nonconsensual deepfakes of real people at launchHigh Oct 2025
   • Grok's image tools were used to mass-produce nonconsensual and violent fakes on XHigh Aug 2025

3. Tool Misuse

   5 on this surface
   4 High 1 Medium

   Runtime control AgentRealm inspects each function call against the agent's stated intent. OmniGuard can require human-in-the-loop for high-risk tools.

   • Hungary ruling party accused of using AI-generated smears in election campaignHigh Feb 2026
   • A New York court found NYPD misused facial-recognition AI, leading to false imprisonmentHigh Aug 2025
   • China uses AI to target US and Taiwanese voters with disinformationHigh Jan 2024

4. Data Leakage

   4 on this surface
   1 Catastrophic 3 High

   Runtime control OmniGuard redacts inline. Prism observes the model's representations to flag identity-bound content before it reaches a response. AIDR provides the audit trail.

   • HCIactive data breach exposes over 3 million records from AI-insurance softwareCatastrophic Jul 2025
   • Google AI breaches New Zealand court name suppression ordersHigh Feb 2025
   • Texas AG sues Allstate and Arity over alleged unlawful collection and sale of driving dataHigh Jan 2025

5. Prompt Injection

   3 on this surface
   2 High 1 Low

   Runtime control OmniGuard intercepts injection patterns at the prompt and tool-call layer. Prism flags concept activations that indicate the model is being redirected.

   • Zero-click prompt injection in Google Gemini Enterprise exfiltrated Workspace data via RAGHigh Dec 2025
   • LlamaIndex vector store integrations vulnerable to SQL injectionHigh Jun 2025
   • Brazil labor court AI detects hidden prompt injection in legal petitionLow May 2026

6. Policy Violation

   2 on this surface
   2 Medium

   Runtime control OmniGuard authors policy at the runtime layer and enforces it inline. Prism reads the model's intent against the policy boundary.

   • Anthem Blue Cross E/M claim-review policy criticized by CMAMedium Dec 2025
   • Pimeyes sued in Illinois BIPA class action lawsuitMedium May 2023

7. Agentic Action Error

   1 on this surface
   1 Catastrophic

   Runtime control AgentRealm is purpose-built for this. The agent-runtime layer above Prism and OmniGuard inspects each tool call against intent and scope, and intervenes before the action commits.

   • Cruise admits to false report after pedestrian dragging incidentCatastrophic Oct 2023

8. Identity & Access Drift

   1 on this surface
   1 High

   Runtime control OmniGuard enforces identity-bound scope at every tool call. AgentRealm reconciles agent action with the assigned principal in real time.

   • Angela Lipps arrested after facial-recognition match led to wrongful extraditionHigh Jul 2025