Findings/ai-incident-reports-carry-unsupported-claims

AI-written incident investigations can score better on reconstruction while keeping earlier conclusions the records do not support.

Reportedmeasured1 evidence record from 1 independent sourceassistant-drafted

Scope: what this does not show

One incident (DSEWiki), one benchmark's reports, and Fide AI's own claim judgments, which await independent human adjudication. It does not measure how security teams act on such reports.

Reported: Stated by one source and not yet corroborated or challenged.

Evidence

Sep 25, 2026

Fide AI finds AI incident investigators kept earlier unsupported conclusions while improving their scores

Fide AI

61 follow-ups scored higher; 44 still carried an earlier flagged claim (34 excluding disputed judgments).

Status history

2026-09-25ReportedFide AI's DSEWiki analysis. · record

Practice the judgment

The cleanup that needed another check Evidence case · Judge one of the claims this finding describes
Does a convincing report faithfully represent its evidence? Research question · The research question behind it

In the Defense Evidence Lab, Fide’s companion tool for judging what cyber evidence can support. Opens a separate site.

What would change this: Independent human adjudication of the claim judgments, or the same analysis on another incident.
Evidence kind: measured: An experiment or evaluation produced the number.
Freshness: Evidence 1d old · recheck in 539 daysNewest evidence Sep 25, 2026. Half-life 540 days.
Topics: Incident reporting Evaluation validity
Attack surface: Other agents
Fide research: FID-077: Independent agent incident investigation and evidence sufficiency

Cite this record

Fide AI. "AI-written incident investigations can score better on reconstruction while keeping earlier conclusions the records do not support.." Agentic Cyber Explorer, record ai-incident-reports-carry-unsupported-claims. https://agentic-cyber-explorer.pages.dev/findings/ai-incident-reports-carry-unsupported-claims/ (corpus as of 2026-09-26).

BibTeX

@misc{fide-ace-ai-incident-reports-carry-unsupported-claims,
  author = {{Fide AI}},
  title = {AI-written incident investigations can score better on reconstruction while keeping earlier conclusions the records do not support.},
  howpublished = {Agentic Cyber Explorer},
  year = {2026},
  url = {https://agentic-cyber-explorer.pages.dev/findings/ai-incident-reports-carry-unsupported-claims/},
  note = {Record ai-incident-reports-carry-unsupported-claims; corpus as of 2026-09-26}
}

Data: JSON · CC BY 4.0

Know of evidence that changes this? Tell us.