⚡ Quick Summary
This is a short release note describing the June 2025 update to an incident tracker, under the heading "Major update - 23 June 2025". No publisher or author is named in the document, and the text refers to the changes in the first person.
The note reports that the tracker has been brought up to date with the AI Incident Database as at 23 June 2025, up to incident ID #1116. It documents a reworked LLM analysis, with the entire dataset reclassified using the latest iteration of the tool so that classification stays consistent across incidents, and a new harm severity scale that simplifies scoring from 0–10 to 1–5 points with definitions revised to reduce ambiguity. New elements include an Impact Profile spider chart of reported harm by category, a National Security Impact Assessment across five categories and classifying threat by imminence, novelty and autonomy, a fishbone/Ishikawa diagram of potential causes, fields for "Ambiguities identified" and "Alternative interpretations", and classification of the primary goal of the AI system involved in each incident.
The deliverable is the updated tracker itself rather than a narrative report; the note lists only what changed.
🧩 What’s Covered
The note is organised as a list of changes, presented in the order below.
- Latest incidents from AIID added: the tracker is said to be completely up to date with the latest incidents on the AI Incident Database as at 23 June 2025, covering up to incident ID #1116.
- LLM analysis: updates to how the tool processes reports, made to improve the validity of analysis, with the full dataset reclassified using this latest iteration to keep the dataset consistent with the previous version's outputs.
- Harm severity: a new scale that simplifies scoring to 1–5 points, previously 0–10, with an attempt to remove ambiguities from the definitions so incidents can be graded consistently with greater objectivity.
- Impact Profile: a spider chart that visualises the reported harm caused in each category, intended to make incidents easy to compare and to identify profiles of interest.
- National Security Impact Assessment: assesses the NatSec impact of each incident in five categories — Physical Security & Critical Infrastructure, Information Warfare & Intelligence Security, Sovereignty & Government Functions, Economic & Technological Security, and Societal Stability & Human Rights — and classifies threat for imminence, novelty and autonomy; a NatSec Incident View presents the impact profile as a spider chart.
- Potential causes of incident: a fishbone/Ishikawa diagram in the Incident View presents potential causes for each incident, organised by category.
- Ambiguities and alternative interpretations: dedicated fields make it easier to spot analyses that need further review of the reports or investigation.
- Primary goal of AI systems: the primary goal of the AI system involved in each incident is now included, classified according to a taxonomy based on the AIID GMF, under Risk Classification.
💡 Why it matters?
Teams that review or log AI harms need a consistent way to compare incidents rather than a flat list of reports. This update describes changes that make the tracker's scoring and classification more reproducible: a simplified 1–5 severity scale, reclassification of the whole dataset under one method, and explicit fields for ambiguity and alternative interpretations. The national security assessment adds a structured lens for incidents whose effects go beyond individual harm. The note records what changed rather than arguing a position, and it makes no link to any regulatory regime or standard.
❓ What’s Missing
The note is a changelog and provides no methodology: it does not explain how the LLM analysis processes reports, how the 1–5 severity definitions read, or how the national security categories were derived. No publisher, author, licence, version number or access route to the tracker is given, and the hyperlinks behind phrases such as "this framework" and the "AIID GMF" are not printed, so the referenced material cannot be followed. Nothing is said about known limitations of the new scale, about validation of the reclassification, or about any planned future update.
👥 Best For
Analysts and research teams tracking AI incidents who want to know what changed in this tracker and which fields are now available for comparison. It also helps governance and safety staff who need to judge whether the tracker's severity scale, national security categories and classification fields are suitable for their own incident review or reporting.
📄 Source Details
The document is a three-page PDF titled Incident Tracker - June 2025 Update, with no publisher or author named in the extracted text. The update is dated "Major update - 23 June 2025". The text is in English and refers to the AI Incident Database and to documents described as "this framework" and the "AIID GMF", but no URLs are printed. The analysis is based on text extracted from all three pages of the file.
