An AI that reports nothing found cannot tell you whether it looked

Jesper E. Thomsen · 9 September 2026

An AI system that reports "no exceptions" writes that sentence with the same confidence whether it searched the whole record or a fraction of it, and the output carries no trace of which. A constraint on the shape of the output holds that fault in check; a rule asking the model to judge its own work does not, and the facts it fails to join need a second pass.

"Nothing found" is the most dangerous sentence an AI system produces, because it reads the same whether the system searched the whole record or a fraction of it.

I run a small fleet of AI agents over my own notes and correspondence, because that is the domain where I can judge the output. The two findings below come from a month of my own corrections. I think both apply to any AI reporting, monitoring or compliance deployment.

An AI "no" is easy to treat as a weaker version of an AI "yes". That gets the shape wrong. A positive claim usually arrives with the thing it found attached. A negative claim arrives with nothing, and the system cannot tell you what it looked at.

Absence reported as fact

Retrieval returns a subset of the corpus. The model treats the subset as the population. It then generates a negative claim ("nothing outstanding", "no matches") with exactly the confidence it gives a positive one, and nothing in the output separates "not in what I retrieved" from "not true".

Five instances in one month, five different agents, each caught by a correction of mine. The agent that watches bills reported no new bills while a bill from the 14th sat unprocessed; my whole correction was "from the 14th?". The agent that tracks correspondence reported ten days of silence on a thread. It was three. A weekly review raised as still open a matter we had already cleared up. None of the five read differently from a run where the claim was true.

The one agent in the fleet I have never corrected, across three runs and roughly sixty findings, is the one whose output is a quotation, a line number, an issue and an action. It cannot say "nothing" without pointing at what it read.

The control case forces a distinction. A rule that asks the model to judge itself ("check whether you have searched everything") does not fire inside the pass. A constraint on the shape of the output ("every negative reads 'not found in [named source]'", "quote the line") holds, because it leaves the model no sentence to write that does not carry its own evidence. Both sit in the prompt; only the second removes the option of an unsupported sentence.

The model does not join its own premises

Working through a board financing note for a company I chair, in a long session with many established facts, the assistant quoted the CEO's own travel gate, then flagged the absence of a pre-agreed exit criterion. Its own text, earlier in the session. There were several more in that session, and in each the missing input was already in its context, written by itself. (A further miss, an option set that did not map onto a structure set out days earlier, I leave out; the structure may not have been in view.)

This failure produces negatives too: "no exit criterion agreed" is an absence reported as fact with the record in front of the model. So the first fault has two causes, and the output-shape constraint reaches one of them. A quote requirement makes the model point at what it read, not read what it wrote.

The mechanism is that the model generates forward and does not re-read its own output as evidence. Each premise is produced, emitted and dropped from the working set. The failure is structural, not informational.

Fourteen standing rules and twenty-five operating principles were in context during that session, and none fired. An instruction of that kind asks the generating process to critique itself while it is generating, and it does not. The same model, given the transcript afterwards, diagnosed the misses precisely. Same model, different mode. The fix is architectural: a second pass, with the finished artefact as input and a clean context. A pass of that kind exists in my fleet, unvalidated.

The second pass carries the first finding one level up. Its usual output on a clean artefact is "nothing found". Unless that sentence is under the same source constraint, naming the artefact and the lines checked, the reader holds a negative with nothing attached, from the pass meant to catch them.

What this means for a board

Two questions worth asking a vendor of AI reporting, monitoring or compliance before signing.

• Show me a negative result. Where in it is the quotation, and the location, of what was read? A list of sources checked does not count; the list is generated by the same system and can be asserted without the search having been done. The demand is for the output shape.
• What reads the output before you do, and is it a separate call? A second pass with the artefact as input and a fresh context is one answer. An instruction in the same call to review the work is not, however many rules it contains.

A separate argument holds that an agentic layer in front of the old stack lowers programme risk in a migration. This is the risk it puts back: the layer reports on the systems behind it, and its negatives are the ones nobody checks.

Where the argument does not hold

The first failure is proportional to how much of the corpus the agent can see in one pass. Where the whole record fits in front of the model, or the agent is constrained to quote, the retrieval cause disappears, though the join cause does not. The claim is about systems that summarise more than they can read in one pass.

The second failure is worse in long sessions under production pressure, and absent or much reduced in retrospective analysis of a closed artefact with one question in front of it. That is the reason to expect a second pass to work. The pass has found misses known to be there; untested is its false-negative rate where nobody knows whether a miss exists, and that is the test: does it surface a join I would otherwise have had to supply, or does it return reformatting? If reformatting, the mechanism is misdiagnosed and the problem is retrieval depth, not reasoning.