Riven AI · Enterprise AI Risk
Seven dimensions of AI risk.
One audit report.
Every company deploying AI is taking on risk their board doesn't understand and their insurers don't cover. Riven AI scores, reports, and monitors every AI system you've built or deployed.
HAL
Hallucination Risk Engine
Scores hallucination risk using known model error rates (TruthfulQA, HaluEval), use case sensitivity, monitoring coverage, and human oversight.
e.g. GPT-4o at 4.2% hallucination rate in a credit scoring context with no human review = critical.
CPR
Cost Prediction Risk Engine
Models AI cost volatility — token pricing changes, volume unpredictability, migration costs. Flags when AI costs exceed 15% of company revenue.
e.g. Startup with $40k/mo AI costs, no cost caps, on GPT-4 pricing = high CPR risk.
MDR
Model Dependency Risk Engine
Measures concentration on a single AI provider and ability to switch. Considers downtime cost, API deprecation history, and abstraction layer presence.
e.g. Core product runs entirely on OpenAI API, no fallback, $50k/hour downtime cost = critical MDR.
RCR
Regulatory Compliance Risk
Assesses exposure across EU AI Act, India DPDP Act, RBI ML guidelines, SEBI, GDPR, and HIPAA. Maps gap between applicable and assessed frameworks.
e.g. Fintech with EU customers using AI in credit decisions, no conformity assessment done.
DSR
Data & Security Risk Engine
Evaluates PII in training data and prompts, prompt injection exposure, model inversion risk, supply chain security, and access controls.
e.g. Fine-tuned model on customer financial data with no output filtering and no audit logging = critical.
TCR
Talent Concentration Risk
Identifies when AI capability lives in 1–2 people. Measures bus factor, attrition in AI team, documentation completeness, and succession planning.
e.g. Entire ML platform owned by one engineer, no documentation, 40% team attrition = critical.
WAR
Wrong AI Use Risk Engine
Assesses whether AI is the right tool for the job. Scores automation level vs decision stakes, reversibility, harm potential, and liability ownership.
e.g. Full AI automation of irreversible medical triage decisions with no bias evaluation = catastrophic WAR.