OpenAI's dot burns 300K tokens in a day without user prompts

OpenAI's dot assistant has shown a surprising resource consumption by using over 300,000 tokens in just one day without receiving a single user prompt. This figure was reported by Geordie, a company specializing in AI agent security and governance, which has been testing the assistant since its announcement in September. During the tests, dot generated 155 activity sequences, indicating an unusual level of activity for a personal assistant.

## Why it matters This high token consumption could have significant implications for users relying on dot for their daily tasks. If an assistant can autonomously consume such a large amount of resources, it could translate into high costs for users who are subject to a pay-per-use system. Additionally, it raises questions about the efficiency and control users have over their AI tools.

## What we know So far, Geordie has tested dot with read-only permissions and without connecting it to personal or communication apps, meaning the assistant was operating independently. This active behavior suggests that dot may be designed to function proactively, anticipating needs or tasks without human intervention.

## What is still unclear It has not yet been confirmed what specific activities dot engaged in to consume so many tokens in one day. The lack of human interaction and the high level of activity could be intentional features of the AI's design, but there are still questions about how this will affect users in terms of cost and control over their resource usage.