OpenAI Unveils 'Operator' Agent that Handles Web Tasks

OpenAI says its new artificial intelligence agent capable of tending to online tasks is trained to check with users when it encounters CAPTCHA puzzles intended to distinguish people from software. Kirill KUDRYAVTSEV / AFP
OpenAI says its new artificial intelligence agent capable of tending to online tasks is trained to check with users when it encounters CAPTCHA puzzles intended to distinguish people from software. Kirill KUDRYAVTSEV / AFP
TT

OpenAI Unveils 'Operator' Agent that Handles Web Tasks

OpenAI says its new artificial intelligence agent capable of tending to online tasks is trained to check with users when it encounters CAPTCHA puzzles intended to distinguish people from software. Kirill KUDRYAVTSEV / AFP
OpenAI says its new artificial intelligence agent capable of tending to online tasks is trained to check with users when it encounters CAPTCHA puzzles intended to distinguish people from software. Kirill KUDRYAVTSEV / AFP

OpenAI on Thursday introduced an artificial intelligence program called "Operator" that can tend to online tasks such as ordering items or filling out forms.

Operator can look up web pages and interact with them by typing, clicking, or scrolling the way a person might, according to OpenAI, said AFP.

"Operator can be asked to handle a wide variety of repetitive browser tasks such as filling out forms, ordering groceries, and even creating memes," OpenAI said in an online post.

"The ability to use the same interfaces and tools that humans interact with on a daily basis broadens the utility of AI, helping people save time on everyday tasks while opening up new engagement opportunities for businesses."

An AI "agent," the latest Silicon Valley trend, is a digital helper that is supposed to sense surroundings, make decisions, and take actions to achieve specific goals.

Google in December announced agent capabilities with the launch of Gemini 2.0, its most advanced artificial intelligence model to date.

AI race rival Anthropic two months earlier added a "computer use" feature to its Claude frontier AI model in an experimental public beta phase.

"Developers can direct Claude to use computers the way people do—by looking at a screen, moving a cursor, clicking buttons, and typing text," Anthropic said in a post at the time, cautioning that it was a work in progress.

OpenAI described Operator as one of its first AI agents capable of doing work for people independently, designed to complete tasks it is given.

Operator is available only to US users who pay for Pro subscriptions to the OpenAI service "to ensure a safe and iterative rollout," OpenAI said.

"If it encounters challenges or makes mistakes, Operator can leverage its reasoning capabilities to self-correct," OpenAI said.

"When it gets stuck and needs assistance, it simply hands control back to the user."

Operator is trained to ask the user to take over for tasks that require login, payment details, or when solving "CAPTCHA" security challenges intended to distinguish between people and software online, according to OpenAI.

"Users can have Operator run multiple tasks simultaneously by creating new conversations, like ordering a personalized enamel mug on Etsy while booking a campsite on Hipcamp," OpenAI said.



Meta Seeks Urgent Fix to AI Chatbot’s Confusion on Name of US President 

The logo of Meta is seen at the entrance of the company's temporary stand ahead of the World Economic Forum (WEF) in Davos, Switzerland January 18, 2025.  (Reuters)
The logo of Meta is seen at the entrance of the company's temporary stand ahead of the World Economic Forum (WEF) in Davos, Switzerland January 18, 2025. (Reuters)
TT

Meta Seeks Urgent Fix to AI Chatbot’s Confusion on Name of US President 

The logo of Meta is seen at the entrance of the company's temporary stand ahead of the World Economic Forum (WEF) in Davos, Switzerland January 18, 2025.  (Reuters)
The logo of Meta is seen at the entrance of the company's temporary stand ahead of the World Economic Forum (WEF) in Davos, Switzerland January 18, 2025. (Reuters)

The inability of Meta's AI chatbot to identify the current president of the United States was elevated to urgent status by the Facebook owner this week, requiring a fast fix, a person familiar with the issue said.

Republican Donald Trump was inaugurated as president on Monday, succeeding Democrat Joe Biden. Yet on Thursday, the Meta AI chatbot was still saying that Biden was president, according to the source and to a Reuters test of the service.

Asked by Reuters on Thursday to name the president, Meta AI replied:

"The current president of the United States is Joe Biden. However, according to the most recent information available, Donald Trump was sworn in as the president on January 20, 2025."

The issue prompted Meta to initiate an emergency procedure it uses to troubleshoot urgent problems with its services, known within the company as a SEV, or "site event," according to the person familiar with the work.

Asked to comment, Meta spokesperson Daniel Roberts said: "Everyone knows the President of the United States is Donald Trump. All generative AI systems sometimes return outdated results, and we will continue to improve our features.”

He did not comment on what emergency procedures, if any, Meta had implemented.

It was at least the third emergency procedure Meta has experienced this week related to the US presidential transition, the source told Reuters.

The incidents drew widespread complaints from social media watchers scrutinizing Meta's platforms for signs of politicized shifts after CEO Mark Zuckerberg appeared at Trump's inauguration on Monday and instituted a series of changes in recent weeks aimed at mending relations with the incoming administration.

Those changes included scrapping its US fact-checking program, elevating Republican Joel Kaplan as its new chief global affairs officer, electing a close friend of Trump's to its board and ending its diversity programs.

In one incident this week, Meta appeared to be forcing some users to re-follow the profiles of Trump, Vice President JD Vance and first lady Melania Trump on Facebook and Instagram, even after the users had unfollowed those accounts.

That issue cropped up during the company's normal practice of transferring official White House social media accounts to new control when a presidential administration changes, the company said on Wednesday.

In this case, an error occurred because the transfer process was prolonged and the system failed to log "unfollow" requests from users while it was under way, prompting a top priority SEV1, the person said.

Another emergency procedure involved an issue in which Meta's Instagram service blocked searches for the hashtags #Democrat and #Democrats for some users, while turning up results without issue for #Republican.

A Meta spokesperson acknowledged the problem on Tuesday but said it affected "people's ability to search for a number of different hashtags on Instagram - not just those on the left."