L
📕 Companion article · 2 of 5

When the model changes under you

silent upgrades · what drifts · the canary file · what to write down

Your instructions did not change and the output did. Nobody told you, nothing broke, and the difference is small enough that you will notice it in about three weeks — usually from somebody else, about a piece that already went out.

Most drift · does not matterThe drift that does · is narrowOne file · catches it in a minute
What you'll learn

What actually moves when a model is replaced

Not everything, and not randomly. The parts that drift are the parts you never specified, which is why the fix is upstream of the drift rather than downstream of it.

  • What drifts — tone, length, hedging, structure you never named.
  • What holds — anything you specified with a test inside it.
  • The canary — one file, run monthly, that tells you in a minute.
  • The record — which version produced the work you already shipped.
Why it matters · where it's tested

Why the panic is misplaced and the risk is real

Drift is not a failure of the tool and not a reason to distrust it; it is what happens to anything you left unspecified, and it was going to happen the first time somebody else did the work too.

Companion to the 25-day courseBuilds on days 6, 11, 24Stands alone

People look for: AI model drift, changed AI output, model update testing, prompt stability, and version records.

See version drift happen

The left column is what people worry about. The right column is what actually happens. The difference between the two rows is whether you ever wrote the thing down.

What drifts, and what does not
The part of your outputDoes it drift?Why
How long the answer runsYes You never gave a length, so it uses its own default and the default moved
How much it hedgesYes, and most visibly Caution is tuned deliberately between versions. Nothing in your ask fixed it
How it opensYes An unspecified opening is a house style, and the house changed
Whether it follows a rule with a test in itRarely A rule that can be marked right or wrong leaves nowhere to drift to
Whether it keeps your named thingsNo, if you said so Keep every proper noun, number and date is checkable, so it holds
Whether the facts are rightNot in the way you fear It was never a reliable source of facts. That did not change, and it is not drift
The canary file

One file, five items, run on the first of the month. Each item is a small ask whose right answer you already know: a passage that must keep three named things, a rule that must be reported and not fixed, a question your standard says to refuse. Run it, read it, and either nothing moved or you have found out in a minute rather than in a client’s email.

What model drift actually changes

Drift appears most clearly in requirements that were never made testable; controlled examples separate meaningful change from ordinary variation.

Drift is unspecified behaviour, arriving late

Everything you did not decide was decided for you, and defaults are not promises. When the default moves you experience it as the tool changing, and what actually happened is that a gap in your instructions became visible.

Rules with tests do not drift

Keep it concise drifts because nothing in it can be marked. No paragraph over sixty words cannot drift, because there is a right answer and the model either meets it or does not. Every rule you tighten is a piece of your work taken out of the weather.

What to write down, once

Which model produced the work, and on what date. One line in the record from day 20, costing nothing. Six months later, when somebody asks why two pieces from the same quarter read differently, that line is the whole answer.

When drift is not the explanation
  • Different source material can change the output even when the model is stable.
  • Sampling creates variation between repeated runs of the same model.
  • A changed template, system instruction, tool, or retrieval source can look like model drift.
Two signals worth watching
  • A known test begins breaking the same requirement repeatedly.
  • Several unrelated workflows shift in the same direction after the same update.

Version drift in real work

The template that stopped fitting

A monthly report began opening with longer summaries after an upgrade. Its required fields still held; only the unspecified introduction moved, so the team added one measurable opening rule.

Two pieces, one quarter, different voices

Two reports used the same brief but different model versions. The record explained the change, and a small style canary showed which uncontrolled defaults had moved.

The refusal that appeared from nowhere

A previously accepted workflow began refusing a harmless transformation. A narrow test confirmed the boundary change, allowing the team to revise that step without rebuilding everything else.

Glossary — the 4 words that unlock it

Drift

What it means
The output changing while your instructions stay the same, because the model behind them changed.
Why it matters
Nothing breaks and nothing warns you, so you find out weeks later from somebody else.
Example
The same prompt returning noticeably more hedging than it did last quarter.
Key question
Which parts of your output did you never actually specify?

Default

What it means
Whatever the tool decides when you did not say. Length, tone, how much it hedges.
Why it matters
Every default is on loan. Drift is the lender calling it back.
Example
You never gave a length, so it used its own, and its own moved.
Key question
If a default changed tomorrow, would you notice or would a reader?

Canary file

What it means
A small set of asks whose right answer you already know, run on a fixed day each month.
Why it matters
It turns a slow surprise into a one-minute check you chose to run.
Example
Five items: a passage that must keep three named things, a rule that must be reported and not fixed.
Key question
What would tell you the model moved, before a client does?

Pinning

What it means
Recording which model produced a piece of work, and on what date.
Why it matters
Six months on it is the whole answer to why two pieces from one quarter read differently.
Example
One line in the delivery record, costing nothing at the time.
Key question
Could you say today which model wrote what you shipped in March?

Questions about model changes

What is model drift in everyday AI-assisted work?
ConceptualWhatcomplexity 2

Model drift is a noticeable change in output after the provider replaces or retunes the model while your instructions and source material stay the same. It often appears in unspecified qualities such as length, tone, structure, hedging, or refusal behaviour rather than in every part of the work.

Why can the same prompt produce different results after an update?
ConceptualWhatcomplexity 2

A prompt fixes only what it states and demonstrates. Everything left open is handled by the model's current defaults, and those defaults can change between versions. Sampling also creates ordinary variation, so compare several controlled outputs before deciding that an update caused the difference.

Can a team completely prevent model drift?
ConceptualWhatcomplexity 2

No, but it can reduce the part that matters. Write important requirements as testable rules, keep a small set of known-input examples, record the model and date used for consequential work, and review the outputs that carry financial, legal, safety, or reputational risk.

What should a model-drift canary file contain?
ConceptualWhatcomplexity 2

Use a handful of short tasks whose acceptable behaviour is already known: preserve named facts, follow a measurable style rule, report rather than repair a defect, and refuse only where the standard requires it. The file detects meaningful movement quickly; it is not a general model benchmark.

When should existing AI-assisted work be rerun after a model change?
ConceptualWhatcomplexity 2

Rerun work only when future outputs depend on the changed behaviour or when the original result must remain reproducible. Published prose usually does not need regeneration. Templates, automations, regulated material, high-volume workflows, and safety-sensitive decisions deserve targeted retesting.

Key takeaways

  • Most model drift affects unspecified defaults.
  • Testable requirements are more stable than adjectives.
  • A small canary file reveals meaningful movement quickly.
  • Record model and date for consequential work.
  • Retest affected workflows, not every published output.

Where this lesson leads

Stable work comes from narrowing uncertainty and recording what produced consequential outputs.
✓Named requirements
✓Canary file
✓Version record
→Targeted retesting
→Workflow monitoring
→Model-independent standards

Keep work stable

Anything you did not specify is on loan, and the lender can call it back.

▶ Try the playground
© Pawan Nayar · LLOS.ai · 2026 — When the model changes under you.Original pedagogy, voice, and design — all rights reserved.