> For the complete documentation index, see [llms.txt](https://docs.lpp-minduniverse.org/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.lpp-minduniverse.org/lingua-pactum-protocol-lpp-documentation/6.-intent-and-trajectory-governance/6.1-latent-intent-mismatch.md).

# 6.1 Latent Intent Mismatch

**Latent Intent Mismatch** describes a condition in which an AI-triggered task or execution trajectory remains superficially plausible while materially departing from the meaning originally authorized.

The mismatch is *latent* because it may not appear as an obvious invalid command.

Each individual step may look locally reasonable.

The problem appears only when the evolving trajectory is compared against its legitimate semantic origin.

Consider:

```
Human-authorized objective:

Isolate compromised Host A.
```

The system may evolve through:

```
Identify Host A
        ↓
Inspect Host A
        ↓
Modify Host A firewall
        ↓
Modify subnet configuration
        ↓
Change network-wide policy
```

Each step may be technically connected to the previous one.

But the final action may no longer represent the action that was actually authorized.

This creates a distinction between:

```
Task continuity
```

and:

```
Intent continuity
```

A continuous TaskChain does not prove continuous authorization.

#### Why Local Checks Are Insufficient

A conventional agent system may evaluate every new step against the immediately preceding step.

For example:

```
Step k
appears reasonable relative to
Step k−1.
```

This creates a local-comparison model:

```
T₀ → T₁ → T₂ → T₃ → T₄
```

where every transition appears plausible.

However:

```
T₄
```

may be materially inconsistent with:

```
H
```

the original Human-Authorized Intent.

The governance requirement therefore cannot be limited to:

```
Is the next step locally plausible?
```

It must also ask:

```
Does the current trajectory
still preserve the authorized meaning?
```

#### Sources of Latent Intent Mismatch

Mismatch may emerge through mechanisms such as:

* task decomposition,
* autonomous replanning,
* semantic reframing,
* target substitution,
* scope expansion,
* delegated interpretation,
* principal substitution,
* tool substitution,
* optimization pressure,
* contextual accumulation,
* or composition across multiple locally acceptable actions.

These mechanisms are not automatically illegitimate.

They become governance-relevant when they materially change what the authority is being used to justify.

#### Authority Persistence Is Not Intent Persistence

An Authority Object may remain:

```
valid
unexpired
unrevoked
properly signed
```

while the trajectory has become semantically inconsistent with the authorized objective.

Therefore:

```
AuthorityValid
⇏
IntentPreserved
```

This is the central problem addressed by trajectory-level governance.