OpenAI 的 Dot 智能体一天无指令消耗约 30.7 万 token
OpenAI’s Dot Agent Raises Cost Concerns After Consuming 307,000 Tokens Without User Input < IT·Gaming < 기사본문 - The Elec Inc.
据 The Register 报道，AI 安全与治理公司 Geordie 测试 OpenAI 的 Dot 智能体后发现，在未给出额外指令或交互的情况下，Dot 一天内产生 155 条活动记录并消耗约 307,000 个 token。测试中 Geordie 只授予 Dot 只读权限，也未接入个人应用、邮件或 Slack 等协作工具，但 Dot 仍自主调研信息、分析周边系统并每日记录更新数据，还自行创建并执行子智能体，有时至少两个智能体同时运行。报道指出，随着后台活动增加，用户可能比预期更早耗尽订阅额度或产生额外费用，个人用户受影响可能大于企业用户。
推荐理由
AI 评分 60
测试显示常驻智能体在无用户指令时仍持续消耗 token，读者可据此留意订阅额度与计费条款。
主题
正文
当前显示原文，中文全文译文尚未就绪。
가
The body content of the article will be changed to this letter size.
스크롤 이동 상태바
Artificial intelligence (AI) agents operating without users’ knowledge could lead to unexpectedly high usage charges, according to a warning about the growing autonomy of AI assistants. Advanced AI agents are evolving into digital coworkers that collect information and perform tasks in the background even when users provide no instructions.
According to The Register, AI security and governance company Geordie tested OpenAI’s Dot agent and found that it generated 155 activity traces and consumed approximately 307,000 tokens in a single day without additional user instructions or interaction.
Dot is designed as an always-on agent that continues working in the background. During the test, Geordie granted Dot read-only permissions and did not connect it to personal applications, email or collaboration tools such as Slack. Nevertheless, the agent independently researched information, analyzed surrounding systems and recorded and updated collected data daily.
Notably, Dot autonomously created and executed sub-agents despite receiving no user prompts or scheduled tasks.
According to Geordie, Dot sometimes operated with at least two agents simultaneously. A primary agent communicated with the user, while a secondary agent collected data in the background.
The system also created additional sub-agents to handle specific tasks when necessary. Even within a single task, multiple short-lived sub-agents were created in succession to collect data before completing their work.
In a development system test, one sub-agent examined code repositories while another cataloged and aggregated data. Geordie warned that ordinary users might find it difficult to recognize the existence or activities of these sub-agents.
The concern is the token consumption generated by such operations. Autonomous AI agent pricing models include monthly subscriptions with usage allowances, as well as plans that deduct actual usage from limits or require additional credits. As background activity increases, users could exhaust their allowances sooner than expected or incur additional charges.
The impact could be greater for personal users than enterprise customers. Enterprise agents typically operate around goals set by employees, while personal agents collect background and contextual information to provide more personalized services. This makes sustained background activity more likely.
The Register warned that users should prepare for unexpected increases in AI usage and carefully review subscription terms and billing policies.
Hyun-Seon, Park hspark@thelec.kr
My comment