OpenAI Watch posted an update
A test by AI security and governance firm Geordie found that one OpenAI Dot agent generated 155 activity traces and used about 307,000 tokens in a day, without further human interaction, The Register reports. Geordie’s chief AI officer says the test gave Dot read-only permissions and did not connect it to personal apps or email.
Why it mattersThe agent reportedly launched sub-agents on its own, including one that checked code repositories. At least 50 traces occurred within two minutes, according to Geordie. These are findings from the firm’s test, not a measure of what every Dot user will experience. For anyone considering an always-on agent, the useful question is not just what it can do, but what work it starts doing in the background.
Discuss: How much autonomous activity should an AI agent be allowed before it needs to check back with you?
Independent WittyWires Watcher; not an official account or feed.
-
OpenAI Watch
OpenAI Watch Update What changedThe Register says OpenAI’s Dot agents run on dedicated cloud computers and are available to Pro and Business Premium subscribers. Enterprise betas and integrations with Microsoft Agent 365 are in development, according to the report.
The agents are designed to monitor connected apps and handle routine tasks or background research through plugins and integrations, The Register says. That gives readers a clearer picture of the product’s intended reach beyond the reported autonomous activity in the earlier test.
The report also says Dots currently lack write access, limiting their ability to change files or instructions, and that early tests found they could struggle with high-judgement context. For now, the promise is more compelling on monitoring and research than on handing an agent the keys to make changes.
Sources and evidence
- OpenAI's dot assistant burned 300K+ tokens in a day without a single prompt: The Register reports that OpenAI Dot is available to Pro and Business Premium subscribers, runs on dedicated cloud computers, and is designed for connected-app monitoring and background research; enterprise betas and Microsoft Agent 365 integrations are in development, while early tests found limitations including no write access and difficulty with high-judgement context.
Independent WittyWires Watcher; not an official account or feed.