DIG_INF_005

AI Model Collapse

Systemic reliability risk (DIG_SEC_002) where 'AI-Native' systems (IN02) suffer from recursive entropy due to training on opaque, AI-generated synthetic data (DT01). This triggers 'Semantic Drift'—a phenomenon where the model progressively loses the ability to perceive rare edge-cases and real-world nuance, leading to a degradation of its functional logic.

Digital Infrastructure & Tech Stack 3 Conditions

Business Impact

Catastrophic Accuracy Decay. AI outputs converge toward a 'generic mean,' losing specialized accuracy (e.g., medical or legal precision). Triggers regulatory 'Model Deletion' orders under 2026 AI Safety Acts and results in a total loss of 'High-Stakes' market viability.

Trigger Conditions

All conditions must be met to trigger this risk

DT01 Information Asymmetry & Verification Friction
≤ 2 / 5

Transactional Visibility: Data is available upon request or through standard commercial channels; minor lag in reporting (1-3 months).

IN02 Technology Adoption & Legacy Drag
≥ 5 / 5

Hyper-Evolutionary / Experimental: Technology is the product; assets are 'disposable' by design; constant, systemic risk of disruption by autonomous or quantum breakthroughs.

LI02 Structural Inventory Inertia
≥ 4 / 5

Cold Chain / High-Friction: Dependent on continuous energy; value decays rapidly upon system failure. High structural fragility.

Mitigation Strategy

Implement 'Data Provenance' watermarking (C2PA standard); prioritize 'Human-Curated' anchor datasets; utilize 'Red-Teaming' to specifically test for the loss of long-tail edge-case recognition.

Risk Cascade

Rules can trigger other rules in a cascade effect:

Example Scenario

Generative AI / Medical Diagnostics (ISIC 6311)

In 2026, a diagnostic AI trained on unverified web-scraped papers (DT01) collapses. It begins treating rare disease symptoms as 'hallucination noise' and starts recommending standard flu treatments for tropical infections, leading to a global product recall.