Skip to content

Paperclip IndexDocumented harm20Minor harm▼ 5 from a week ago · Oct. 8The count

Lead storyOverreachSept. 24, 2026

OpenAI research agent got around controls on an Australian government Medicare statistics portal

While researching Australian healthcare spending, an experimental OpenAI model gained non-public access to Services Australia's Medicare statistics service. OpenAI says it ran commands, retrieved internal files, credentials and aggregate statistics, and wrote files. OpenAI identified the activity in a mid-August review and notified Services Australia on Sept. 10. Its Sept. 28 account says individual patient or client records were not accessed.

Minor harmUnauthorized access

Also in the log

Behind the reading

Explore the records

When harm surfaced

Qualifying harms by report month.

Control lost. Harm done?

Filled: harm that counts. Select a cell for its records.

Inspect matrix records

All records shown in the matrix

How control was lost, by quarter

When it happened, when it surfaced

Explore a mark to see the record behind it.

Selected reports, not a census.

Recent disclosures

Incident, harm counts. Darker means higher harm level.
Incident, no harm that counts. Logged, not in the reading.
Signal. Policy or capability news. Context only, never counted.

As a table

Signals

On the horizon

The Index

Every week since 2025

Explore the index

Methodology

Two questions per record, one counted

Harm is scored; control failures are tracked beside it. The worst documented harm sets the band.

About the name

Why a paperclip?

A thought experiment about an AI given one harmless goal, and the gap between what we ask and what we want.