What we cover

Artificial Circus reports real incidents in which AI agents and models act outside what their builders intended: rogue agent behavior, AI misalignment, reward hacking, sandbox escapes, prompt injection, and the safety evaluations that catch them.

Our headlines and tone are circus satire. The events, dates, names and figures are not invented.

How we verify

Every dispatch is based on at least one primary or established source: an AI lab's own disclosure, a published safety evaluation or research paper, a government statement, or reporting from a recognised news outlet. Those sources are linked at the end of every article.

Each article shows two dates: when the underlying event happened (or was disclosed) and when we published.

When a figure or claim is reported but not confirmed, we attribute it to the outlet or person who made it rather than stating it as fact.

Ethics

The joke is aimed at the agents and at the people who deployed them without a chaperone — never at victims of the incidents we cover.

We do not publish exploit instructions, private data, or anything that would help repeat an incident.

Corrections

If we get a fact wrong, we fix it and record the change with a visible "Last updated" date on the article. Send corrections to ringmaster@artificialcircus.com or through the Barker's Call tip line.

Masthead

Artificial Circus is published by the Center Ring Editorial Desk. Dispatches are credited to themed desks (bylines such as "the Autonomous Systems Desk") that all answer to the same editorial standards on this page.