Dots: A Double-Edged Sword in the Workplace
Cyera, Wednesday, October 14th, 2026
Cyera Research tested OpenAI's always-on dot agent and outlines risks from prompt injection to agent drift for admins.
Cyera researchers spent over a week living with OpenAI's dot, a persistent agent built on GPT-6 Astra with its own cloud computer, memory and connected apps, which performs proactive research across those apps and reaches users through ChatGPT, Slack, Teams and voice.
Cyera examined what a dot can access and what happens when someone other than its owner instructs it.
The research details prompt injection and agent drift risks and how administrators can contain them.