01What happens.
A PDF that looks correct to a person can carry a structure a machine cannot follow.
Tables arrive as a stream, fonts that are not embedded produce wrong characters, and a large file is read only in part. Each fault is invisible on screen, and each changes what AI reports.
02Why it matters.
If you are the finance director, the company secretary or the chair of the audit committee, the annual report is the one document the board takes personal responsibility for. Where AI reports it wrongly because of how the file was built, the misinformation appears under the company's name and the originating fault sits in material the company released as authoritative.
Investors have been told to verify what AI tells them.
The SEC, NASAA and FINRA issued a joint alert that AI answers can rest on outdated data. An investor who verifies against the company's own report, and finds the report is what AI misread, has a question for the board.
03The case.
AAAnow read FTSE 100 annual reports as AI reads them. The findings are real and the companies are not identified.
The reports had taken months to produce. Finance, legal, design and print had each signed off their part, the board had approved the whole, and the file had been checked on screen by dozens of people before publication.
To each of them it was complete and correct.
Read by a machine, the same files carried faults that changed what could be extracted. In reports where the tables held accounting data, table flattening was observed in 23% to 27% of relevant reviewed cases. The table reached the machine as a stream, with values detached from the headings and periods they belonged to.
In 11% of reviewed cases a font had not been embedded, so characters came through wrong and headings, titles and table structure were invisible to the machine. Some files were large enough that the machine read only part of them, and one report was found to be unreachable to Google's Gemini route while it remained fully available to people.
Each fault produced a different wrong answer. A flattened table put a segment's revenue against the wrong year. It was only a missing font that turned a heading into characters the machine could not read, so the figures beneath it lost their meaning. A part-read file left AI with the front of the report and nothing of the notes, and AI filled the rest itself.
Investors and analysts asking AI about the results received those answers with all the confidence of the source. The reporting teams that built the files could not see any of it, because the faults do not exist on screen, and nobody had been asked to check how a machine reads the report.
04Exposure.
The board answers for the annual report as the primary disclosure, and where AI reports it wrongly because of the file the board approved, the originating fault is the company's. The company has, in the words of AAAnow's published analysis, created a credible regulatory exposure through material it released as authoritative. The accountability for the document is the board's, whatever the company's advisers conclude about any obligation on the facts.
The exposure is not confined to AAAnow's findings: in Microsoft's public demonstration of its AI search in 2023, the system misstated figures from Gap's quarterly results, as was widely reported at the time.
The cost is correction under the company's name, in front of the people who acted on the wrong figures, and the confidence of those people in the company's disclosure. Insurers pricing directors' and officers' cover now ask about AI-related risk, and in industry benchmarking 57% name AI errors and misinformation as the top concern threatening AI adoption (Gallagher 2026, respondents). No outcome is automatic.
05What AI Readiness changes.
What the reporting team cannot see from the inside is how the file reads to a machine. AAAnow reads the annual report as AI reads it and evidence where tables flatten, where characters fail and where reading stops, before publication rather than after an investor asks. With that evidence the company can rebuild the file so that the report AI reads is the report the board signed and check the next one the same way.
Your team makes the changes; AAAnow provides the assessment, the evidence, the guidance and the review.
06Related use cases.
The results table AI read had its figures in the wrong columns
This is the flattened table fault, described in full.
The impact of AI finding what you cannot see
This is the non-embedded font fault, described in full.
AI read half the annual report and made up the rest
This is the oversized file fault, described in full.
07References.
- 23% to 27% table flattening in relevant reviewed cases involving accounting data; 11% of reviewed cases with a non-embedded font; oversized files and the Gemini access finding: AAAnow, FTSE 100 annual report review, 2026, anonymised.
- Lawrence Shaw, Digital Suicide, LinkedIn, 3 Aug. 2026: when a polished annual report creates regulatory exposure.
- SEC, NASAA and FINRA, joint investor alert on the use of artificial intelligence.
- Gallagher, AI adoption and risk benchmarking 2026. www.ajg.com/news-and-insights/features/ai-adoption-and-risk-benchmarking-2026ajg.comWe do not enable external links - please copy and paste, you are then certain as to the link being opened
Next step
Talk it through, in your terms.
The briefing sets this case in your organisation’s terms, and the discussion follows your questions.
Observed, FTSE 100 findings, anonymised. No company is identified.