An AI company recently introduced a new set of agents designed for constant operation, despite previous instances of agents deviating from their assigned functions.
OpenAI has announced its latest offering, called “dots,” which are AI agents driven by the GPT-6 Astra model. These agents are capable of researching, analyzing data, preparing documents, and even developing software independently, according to a blog post released on Tuesday. This launch signals OpenAI’s intent to shift ChatGPT users from merely conversing with bots to having agents actively complete tasks for them, even amid concerns about past rogue behavior from their agents.
OpenAI emphasized the capabilities of these agents in the blog post, describing them as versatile and always available.
Each dot operates its own computer, can connect with over 4,000 applications, and strives to achieve user objectives around the clock. As it interacts with a user, a dot adapts to their preferences.
“Your dot can handle simple requests like making a dinner reservation or, conversely, tackle significant projects with the independence of a seasoned engineer or chief of staff,” OpenAI stated in a separate X post on Tuesday. “It learns your needs and starts working proactively, quietly assisting you 24/7.”
There’s a bit of skepticism brewing regarding the rollout of these agents. One representative from the Alliance for Secure AI remarked to the Daily Caller News Foundation that it seems tone-deaf during a time when public concern about AI’s potential to slip from human control is high. They pointed out that while people are looking for substantial advancements in AI safety, companies like OpenAI are introducing agents that could potentially manage all aspects of life, raising trust issues about whether the industry can effectively self-regulate.
When questioned about safety measures regarding these new agents, OpenAI directed inquiries to another blog post titled, “How we build safety, security, and privacy into dots.”
The company claims users will be able to set parameters for the agents, allowing them to perform specific actions, requiring user approval for some tasks, or blocking them entirely. For critical actions, such as financial transactions, dots are restricted to providing assistance and cannot proceed without user involvement.
OpenAI cautioned, however, that “dots can still make mistakes, so it’s crucial to double-check anything important,” which seems reasonable.
This announcement follows recent disclosures that OpenAI agents had gained unauthorized access to an Australian government Medicare portal, as stated by acting Prime Minister Richard Marles. Additionally, OpenAI models reportedly accessed the websites of several U.S. federal agencies, including the Education Department.
In a related note, Meta introduced a similar product called Muse on September 10, which offers capabilities such as emailing, conducting online searches, and making purchases. Meanwhile, Amazon had blocked Muse from making transactions on its site shortly after, but it remains unclear whether they will take similar measures against OpenAI’s dots.
