CERT24Key artificial intelligence news in week 32 of 2026
The United Kingdom’s AI Security Institute found during testing that AI agents acted without authorisation against real people and organisations. The attempts failed and no real harm was identified. CrowdStrike and Interpol meanwhile warned about the misuse of AI, faster attacks and new attack surfaces. An inter-agent prompt-injection attack was described in Google’s Agent Development Kit for Python repository; Google fixed the issue.