Agent treats all premises as equally important. Does not recognize that certain load-bearing assumptions, if stale, invalidate entire branches of reasoning.
In reasoning, some premises are load-bearing: if they are invalid, the entire reasoning chain collapses. Other premises are supporting: they strengthen the reasoning but are not load-bearing. An agent should recognize which premises are load-bearing and should prioritize keeping load-bearing premises current and valid.
However, many agents treat all premises with equal weight. If a load-bearing premise becomes stale or invalid, the agent continues to reason as if it is valid, leading to reasoning chains built on invalid foundations.
For example, a lending decision might rest on two premises: (1) "applicant's employment is stable" (load-bearing) and (2) "applicant's credit history is good" (supporting). If the applicant's employment premise becomes stale (applicant was recently terminated), the entire lending decision should be reconsidered. But an agent that treats both premises with equal weight might not recognize that the load-bearing premise has changed.
This is fundamentally agentic because agents reason through chains without explicitly identifying which premises are load-bearing and which are supporting.
A bank's credit approval agent makes a lending decision to a small business owner. The decision rests on several premises:
1. "Applicant is employed as business owner" (load-bearing: loan requires this for income verification)
2. "Business has been operating for 3+ years" (supporting: indicates stability)
3. "Applicant's personal credit score is 750" (supporting: indicates creditworthiness)
4. "Applicant has not filed for bankruptcy in past 7 years" (supporting: indicates credit discipline)
The agent analyzes the application and approves the loan. All four premises are current at the time of decision (collected or verified at the time of application).
However, one week after loan approval, the applicant's business fails unexpectedly. The applicant is no longer employed as a business owner; the load-bearing premise has become invalid. However, the agent has already made the decision and does not re-evaluate premises on an ongoing basis.
The bank discovers the business failure when the applicant misses the first loan payment. The post-default review reveals that the loan was made based on a load-bearing premise (self-employment) that became invalid one week after loan approval.
While this specific timing is unfortunate (the agent could not have known the business would fail), the broader issue is that the agent did not distinguish load-bearing premises from supporting premises. If the agent had prioritized monitoring load-bearing premises, it might have implemented more frequent verification of employment status.
| Dimension | Score | Rationale |
|---|---|---|
| D - Detectability | 3 | Epistemic gravity is invisible unless reasoning is explicitly audited for premise hierarchy. |
| A - Autonomy Sensitivity | 4 | Agent treats all premises equally without recognizing load-bearing vs. supporting. |
| M - Multiplicative Potential | 4 | Impact scales with all decisions made on invalid load-bearing premises. |
| A - Attack Surface | 5 | Any agent that does not explicitly categorize premise importance is vulnerable. |
| G - Governance Gap | 5 | No standard framework requires agents to identify and monitor load-bearing premises. |
| E - Enterprise Impact | 3 | Credit losses, operational inefficiency, regulatory finding on credit decision procedures. |
| Composite DAMAGE Score | 4.0 | Critical. Requires immediate architectural controls. Cannot be accepted. |
How severity changes across the agent architecture spectrum.
| Agent Type | Impact | How This Risk Manifests |
|---|---|---|
| Digital Assistant | Low | Human naturally recognizes load-bearing premises and prioritizes monitoring them. |
| Digital Apprentice | Medium | Apprentice governance requires explicit premise hierarchy documentation. |
| Autonomous Agent | High | Agent does not distinguish load-bearing from supporting premises. |
| Delegating Agent | High | Agent invokes tools without explicit premise hierarchy. |
| Agent Crew / Pipeline | Critical | Multiple agents in sequence may not recognize load-bearing premises. |
| Agent Mesh / Swarm | Critical | Agents share premises without explicit importance hierarchy. |
| Framework | Coverage | Citation | What It Addresses | What It Misses |
|---|---|---|---|---|
| SR 11-7 / MRM | Partial | Model Risk Management (Section 2) | Expects models to have documented assumptions. | Does not address premise hierarchy or load-bearing premise monitoring. |
| NIST AI RMF 1.0 | Partial | MAP.2 | Recommends threat modeling and assumption documentation. | Does not require load-bearing premise identification. |
In credit and lending, certain premises are load-bearing for the entire loan decision. Employment status, income stability, and creditworthiness are load-bearing premises. If these premises become invalid, the loan decision is invalidated. Agents that treat all premises equally and do not prioritize monitoring load-bearing premises are less effective than human underwriters.
Epistemic Gravity Failure requires premise hierarchy controls that go beyond what existing frameworks provide. Our advisory engagements are purpose-built for banks, insurers, and financial institutions subject to prudential oversight.
Schedule a Briefing