OpenAI simply launched its new ChatGPT Agent that may make as many as 1 sophisticated cupcake order per hour, however even Sam Altman says you in all probability shouldn't belief it for 'high-stakes makes use of'
OpenAI launched ChatGPT Agent on Thursday, its newest effort within the industry-wide pursuit to show AI right into a worthwhile enterprise—not only one that eats buyers’ billions. In its announcement weblog, OpenAI says its Agent “can now do give you the results you want utilizing its personal pc,” however CEO Sam Altman warns that the rollout presents unpredictable dangers.
AI brokers are machine studying instruments meant to carry out complicated, multi-step duties, and so they’ve been the newest landmark within the AI arms race for rivals like Google and Microsoft. In prerelease demos for Wired and The Verge, OpenAI presenters used ChatGPT Agent to automate calendar planning and creating monetary displays.
By mixing its earlier Operator and deep analysis agentic fashions, OpenAI says Agent can carry out “complicated duties from begin to end.” In accordance with OpenAI spokespeople, these duties usually take Agent 10 or quarter-hour, whereas extra sophisticated assignments take the device longer to finish.
OpenAI analysis lead Lisa Fulford informed Wired that she used Agent to order “lots of cupcakes,” which took the device about an hour, as a result of she was very particular in regards to the cupcakes.
“It was simpler than me doing it myself,” Fulford stated, “as a result of I did not need to do it.”
Whereas the potential cupcake timesavings alone are functionally infinite, Altman took to X as we speak to warn that utilizing Agent might current some appreciable risks—the extent of which OpenAI is outwardly content material to let its customers determine.
“I’d clarify this to my circle of relatives as leading edge and experimental; an opportunity to attempt the longer term,” Altman stated, “however not one thing I’d but use for high-stakes makes use of or with lots of private info till now we have an opportunity to check and enhance it within the wild.”
In the present day we launched a brand new product known as ChatGPT Agent.Agent represents a brand new degree of functionality for AI techniques and may accomplish some exceptional, complicated duties for you utilizing its personal pc. It combines the spirit of Deep Analysis and Operator, however is extra highly effective than that…July 17, 2025
Inspiring the alternative of confidence, Altman stated that “unhealthy actors could attempt to ‘trick’ customers’ AI brokers into giving personal info they should not and take actions they should not, in methods we will not predict.” I am unsure what utility placing these quote marks round “trick” in his X put up supplies, however I am admittedly not a tech visionary.
Altman stated giving Agent greater than “the minimal entry required” or giving it a carte blanche license to reply all of your emails no questions requested might expose vulnerabilities for malicious actors to take advantage of. To mitigate these hazards, Altman stated OpenAI has “constructed lots of safeguards and warnings,” however notes that the corporate “cannot anticipate all the things.”
“Within the spirit of iterative deployment, we’re going to warn customers closely and provides customers freedom to take actions fastidiously in the event that they need to,” Altman stated.
Personally, I’d encourage any customers to need to. Only a few weeks in the past, the CEO of encrypted messaging app Sign warned in regards to the safety dangers of ‘agentic’ AI and the way a lot private knowledge they’re going to require entry to. “There isn’t any mannequin to do this encrypted,” Meredith Whittaker stated in an interview at SXSW.
Price a watch:
Head of Sign, Meredith Whittaker, on so-called “agentic AI” and the distinction between the way it’s described within the advertising and marketing and what entry and management it might really require to work as marketed.— @keithfitzgerald.bsky.social (@keithfitzgerald.bsky.social.bsky.social) 2025-07-17T21:45:54.414Z
“There is a profound concern with safety and privateness that’s haunting this type of hype round brokers, and that’s finally threatening to interrupt the blood-brain barrier between the appliance layer and the OS participant by conjoining all these separate providers, muddying their knowledge,” Whittaker continued. “As a result of hey, the agent’s obtained to get in, textual content your pals, pull the info out of your texts and summarize that in order that your mind can sit in a jar and you are not doing any of that your self.”
OpenAI says Agent is educated to require permission earlier than “taking actions with real-world penalties, like making a purchase order”—which is sweet to know, however I can not assist however marvel how slim the definition of “real-world penalties” is there. Are there real-world penalties if Agent plans a shitty date itinerary?
Likewise, sure “crucial duties” like sending emails would require the consumer to actively supervise Agent’s work. It is also educated to refuse probably catastrophic duties like financial institution transfers or different monetary actions.
OpenAI additionally makes certain to notice that it does not “have definitive proof that the mannequin might meaningfully assist a novice create extreme organic hurt.” So, you understand. That is good.
ChatGPT Agent is accessible now for Professional customers, whereas Plus and Crew customers will obtain entry within the subsequent few days. I am certain it will be superb.