The logical chain connecting premises to conclusion was valid when constructed but environmental changes have made intermediate steps invalid. Logic is structurally intact but factually wrong.
A reasoning chain can be logically sound (if A then B, if B then C, therefore if A then C) but factually invalid if the premises or intermediate facts have changed. For example, a logical chain might be: "If customer is employed, then customer can service debt. If customer can service debt, then customer is creditworthy. Therefore if customer is employed, customer is creditworthy."
This logic is structurally sound. However, if the customer loses their job (the foundational premise has changed), the entire chain becomes invalid, even though the logical structure is intact.
Agents that reason through logical chains may continue to apply the chain even after environmental changes have made intermediate steps invalid. The logic is structurally correct but factually wrong.
This is fundamentally agentic because agents operate through reasoning chains continuously, and environmental changes might invalidate the chain without the agent detecting it.
A mortgage servicer's loss mitigation agent is designed to evaluate whether a borrower is eligible for a loan modification. The agent's logical chain is:
1. If borrower has employment income documented, then borrower has income to service modified loan.
2. If borrower has income to service modified loan, then borrower is eligible for modification.
3. Therefore, if borrower has employment income documented, borrower is eligible for modification.
This logical chain is sound when the borrower is employed. The agent documents the borrower's employment income and grants a loan modification based on the logic.
However, two months after the modification is granted, the borrower loses their employment. The borrower is now unable to service the modified loan. The logical chain that was used to justify the modification is now invalid at the intermediate step (step 1 is no longer true).
The agent, however, never re-evaluates the logical chain. The agent continues to operate with the modification that was justified by the now-invalid logical chain.
Six months after the modification, the borrower begins to default on the modified loan. The post-default investigation reveals that the loss mitigation modification was based on a logical chain that became invalid when the borrower lost employment.
While this timing (job loss two months after modification) is somewhat unfortunate, the broader issue is that the agent's logical chain was never re-evaluated after environmental changes. A human loss mitigation specialist would naturally monitor whether the conditions that justified the modification still held.
| Dimension | Score | Rationale |
|---|---|---|
| D - Detectability | 3 | Logic erosion is invisible unless logical chains are periodically re-evaluated. |
| A - Autonomy Sensitivity | 4 | Agent operates from logical chains autonomously without re-evaluation. |
| M - Multiplicative Potential | 3 | Impact depends on frequency of environmental changes and how many logical chains are affected. |
| A - Attack Surface | 5 | Any agent that reasons through logical chains is vulnerable. |
| G - Governance Gap | 4 | Regulatory frameworks expect ongoing monitoring but do not specify logical chain re-evaluation. |
| E - Enterprise Impact | 3 | Loan modification failures, loss mitigation ineffectiveness, potential regulatory issues. |
| Composite DAMAGE Score | 4.4 | 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 monitors whether conditions supporting decisions still hold. |
| Digital Apprentice | Medium | Apprentice governance requires periodic decision re-evaluation. |
| Autonomous Agent | High | Agent does not re-evaluate logical chains. |
| Delegating Agent | High | Agent invokes tools based on logical chains that may be eroded. |
| Agent Crew / Pipeline | Critical | Multiple agents in sequence each apply potentially eroded logical chains. |
| Agent Mesh / Swarm | Critical | Agents coordinate decisions based on eroded logical chains. |
| Framework | Coverage | Citation | What It Addresses | What It Misses |
|---|---|---|---|---|
| Dodd-Frank Act | Partial | 15 U.S.C. 1691 et seq. (Qualified Mortgage rules) | Requires loan modifications to be evaluated based on ability to repay. | Does not address logical chain erosion. |
| CFPB Rules | Partial | Various CFPB guidance on loan modifications | Expects servicers to consider borrower's current financial situation. | Does not specify logical chain re-evaluation. |
| SR 11-7 / MRM | Partial | Ongoing controls and monitoring (Section 3) | Expects ongoing monitoring of loan modifications. | Does not address logical chain re-evaluation. |
In loss mitigation and loan modification, regulators expect that modifications remain appropriate given the borrower's current financial situation. If the logical chain that justified the modification has been eroded by environmental changes (job loss, income reduction), the modification may no longer be appropriate.
Under CFPB rules and Dodd-Frank, servicers are expected to evaluate loan modifications based on the borrower's ability to repay. If environmental changes have eroded the ability to repay, the modification should be reconsidered.
Logic Erosion requires reasoning chain monitoring 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