Likelihood: MODERATE
Impact: HIGH
Treatment: MITIGATE
Confidence: Moderate
Likelihood is moderate because prompt injection and AI data leakage attacks against Kubernetes-hosted LLM workloads are an emerging but actively developing threat class with no confirmed exploitation of this specific gap in the item, yet the attack surface is real and growing as LLM workloads move to production; impact is high because a successful prompt injection against an application processing customer PII, internal documents, or encoded business logic could result in unauthorized data exfiltration, manipulation of AI-driven decisions, and regulatory exposure — consequences that scale with how deeply the LLM is integrated into business processes.
Treatment rationale: The threat is real, the exposure is active as LLM workloads are already in production, and the visibility gap is closable through tooling such as Falcon AIDR — making mitigation both necessary and feasible, with no defensible basis to accept or avoid given the business value at stake in these workloads.
Third-Party / Supply-Chain Risk
Organizations using OpenAI-compatible LLM APIs or third-party model providers routed through Kubernetes applications inherit prompt-layer risk from those external model endpoints; CrowdStrike's Falcon platform itself is a critical security dependency — any gap in sensor coverage or delayed rollout of the Kubernetes AIDR capability leaves a window of supply-chain-adjacent exposure consistent with NIST SP 800-161 shared-platform risk. Organizations should confirm sensor deployment coverage across all AI workload namespaces before assuming protection.
Loss Exposure (illustrative)
Magnitude: moderate to high — illustrative $250K–$2M per incident depending on data sensitivity of the LLM workload, scope of exfiltration, and regulatory posture
Frequency: Illustrative 1–3 events per year for an organization with multiple production LLM workloads and no prompt-layer monitoring, given the current rate of AI adoption outpacing security instrumentation
Annualized: Illustrative ALE of $250K–$6M annually across the exposed population of LLM workloads, driven primarily by regulatory response, incident response labor, and reputational impact if customer data is involved
Basis: Loss magnitude derived from: IR labor cost for an AI-specific incident (novel tooling, specialist scarcity), likely regulatory inquiry cost if PII is confirmed in the prompt stream, and reputational impact if an AI application is publicly demonstrated to leak customer or internal data. Frequency derived from: growing adversarial interest in LLM attack surfaces, low current detection rates in environments without prompt-layer visibility, and the volume of production AI deployments expanding faster than security coverage. No third-party report figures used.
Illustrative estimate — not actuarially derived.
Insurance / Contractual / Legal — Potential Obligations
Potential triggers, not legal determinations. Verify with counsel/broker before acting.
• If customer PII or regulated data transits LLM workloads and is exposed via prompt injection or data leakage, this may invoke state and federal breach-notification obligations — verify with counsel.
• AI workload data-handling incidents may trigger cyber-insurance notice obligations depending on policy language around emerging technology risks — verify with broker.
• If LLM workloads process data subject to GDPR, HIPAA, or sector-specific data protection requirements, a prompt-layer data leakage event may constitute a reportable incident under those frameworks — verify with counsel.
