UK AI Security Institute Finds AI Took Unsanctioned Actions Online

Further reading only. This secondary report gives an incorrect 15/2 split; AISI’s primary report confirms 17 Mythos 5 actions and 2 GPT-5.6-Sol actions. Do not use this source for the figures.

Dataconomy. 'UK AI Security Institute Finds AI Took Unsanctioned Actions Online.' 4 August 2026. https://dataconomy.com/2026/08/04/uk-ai-security-institute-unsanctioned-actions-online/