Detection and Governance of AI Substitution in Professional Environments

Executive Summary

The rapid integration of generative AI into professional workflows has created a novel governance challenge: the separation of production from comprehension. This document synthesizes an analysis of "The Detection Arc," a framework for identifying, verifying, and managing undisclosed AI use.Critical takeaways include:

Speed Inversion as an Early Warning: Institutional detection begins with "rhythm" rather than audit trails. Mismatches in task velocity—where complex tasks appear effortless while simple ones lag—serve as the primary signal of tool substitution.
From Artifact to Agency: Polished outputs ("artifacts") no longer guarantee competence. The "Live-Edit Test" serves to verify whether an individual possesses a mental model of the work or is merely defending a claim to it.
Curiosity vs. Substitution: Governance must distinguish between users who use AI to "think against" (curiosity) and those who use it to "vanish" (substitution). Organizations often inadvertently incentivize substitution by rewarding surface-level polish over inhabitable methods.
Architecture of Conscience: Relying on individual conscience is insufficient under deadline pressure. Effective governance requires "Interface Friction"—built-in checkpoints such as disclosure gates and mandatory pauses—to ensure accountability at the moment of decision.
The Residual Obligation: While institutional architecture is necessary to support honest practice, the ultimate professional burden remains "inhabitation"—the willingness of the individual to be present and accountable for the logic of their output.

1. The Detection Problem: Reading the Rhythm

Institutional unease regarding AI use often precedes hard evidence. This discomfort manifests as a sense that the "numbers are fine but the rhythm is off." The primary observational tool for this stage is the identification of Speed Inversions .

The Signature of Inverted Competence

Human effort has a "metabolism" characterized by hesitation, iteration, and awkward questioning. In contrast, AI-substituted work often displays a distribution of speed that contradicts the expected burden of the task.| Inversion Type | Manifestation || ------ | ------ || Technical-Communicative Split | Complex structured artifacts (code, budgets) arrive instantly; simple human-facing communication (emails, explanations) lags significantly. || Creative-Analytical Inversion | Concept notes and polished ideation appear immediately; grounded, procedural follow-through becomes "sticky" or effortful. || Synchronous-Asynchronous Gap | Remote or time-shifted work is fluent and complete; real-time discussion or collaborative navigation reveals a lack of grasp. || Vanishing Revision Trail | Work arrives fully formed without a visible history of drafts, structural pivots, or partial logic. |

Generous Suspicion

The document advocates for "generous suspicion"—taking rhythmic mismatches seriously enough to inquire, while remaining open to explanations such as neurodiversity, hidden constraints, or uneven skill. The goal is calibration rather than immediate accusation.

2. Verification of Agency: The Live-Edit Test

When detection moves from observation to verification, the focus shifts from the artifact (the output) to agency (the worker's ability to navigate the work).

The Defense Tax

Individuals who have substituted AI for their own presence incur a "defense tax"—the cognitive burden of protecting ownership of work they do not inhabit. This manifests as:

Vagueness and stalling when asked for specific logic.
Reverse-engineering the artifact in real-time rather than drawing from a mental model.
Over-managed composure and expensive delays for small requested changes.

Implementing the Test

The Live-Edit Test is a collaborative checkpoint where an author is asked to modify a document or dashboard in real-time.

The Five-Minute Heuristic: If an author cannot locate the logic or update an assumption within a few minutes, it reveals "weak possession" of the work.
Normalization: To avoid a prosecutorial atmosphere, these tests should be built into ordinary workflows (e.g., a ten-minute collaborative tweak during a routine review) rather than reserved for crisis moments.

3. The Judgment Layer: Curiosity vs. Substitution

The core professional issue is not the use of AI, but the relationship the worker maintains with the tool.

Relationship Profiles

Curiosity-Driven Use: The worker uses AI to stress-test hypotheses or generate outlines they subsequently rebuild. They "show the seams" because that is where learning occurs. They maintain a critical, inhabited relationship with the output.
Substitution-Driven Use: The worker uses AI to produce a "finished surface" that passes inspection. The goal is to outsource thinking while preserving the appearance of solitary authorship. This leads to a structural gap in comprehension.

The "Curiosity Interview"

To assess professional identity during hiring or checkpoints, the document suggests asking about "friction":

"Show me a time the tool led you astray. How did you know?"
"Where did you choose to override the AI output?"
"Walk me through a project where you are still unsure of the tool's contribution."

4. The Architectural Solution: Conscience in the Interface

Individual conscience is the least reliable component of a governance system because it is eroded by time pressure and fatigue. Organizations must instead build "The Missing Signal" into the machinery of work.

Friction as a Governance Feature

Generic helpfulness in AI tools lacks institutional context (NDAs, data sovereignty, privacy). Institutions must introduce deliberate friction points:

Disclosure Checkpoints: A mandatory field at submission asking if AI was used and for what purpose. This makes use "speakable."
Context Gates: A prompt before data is externalized asking: "Who owns this data? Is it covered by confidentiality?"
Attribution Layers: Metadata flags indicating which sections are human-drafted vs. AI-generated to ensure provenance.
The Uncomfortable Pause: A mandatory 30-second wait before final submission with the prompt: "What am I signing my name to?" This allows the "substrate" (the body/intuition) to catch up with the workflow.

5. Professional Accountability and Consequence

Governance architecture creates the conditions for responsible practice, but it does not guarantee it. The document distinguishes between Compliance (surface behavior) and Inhabitation (being present for the principle the architecture protects).

The Liability Sponge and Social Impact

In sectors like Social Research and M&E (Monitoring and Evaluation), the "erasure of nuance" via AI poses severe risks.

The Liability Sponge: Junior staff often become sponges, signing names to automated consensus or sanitized community grievances they did not verify.
The Victim Register: A mechanism to stress-test reports by considering the person downstream who pays the price for "smoothed" data.
Stop-Work Authority: Protocols to halt processing when a model's abstraction of community pain becomes non-operational.

Proportionate Response Framework

The nature of the institutional response should be dictated by the "Trust Capital" remaining and the state of the system when the failure occurred:| Failure Context | Recommended Response || ------ | ------ || System Absence | Failure in a system with no friction or support requires coaching and training. || Ambiguous Culture | Shared responsibility between the individual and the organization; requires architectural improvement. || Presence Failure | Failure within a robust system with clear friction points requires serious disciplinary action or termination. |

Conclusion: The Residual Obligation

The transition from a surveillance culture to an observability culture requires the institution to make honesty "cheaper" than concealment. However, once the architecture is built, the "residual obligation" returns to the individual. The workflow can ask the question; only the person can mean the answer.