Welcome to Eye on AI! In today’s edition: Trump signs executive order on AI education…California Bar discloses AI was used to develop some questions in problem-plagued February exam…Anthropic’s chief security officer says AI ‘employees’ are coming within a year…Measuring the AI revolution

Google executive Sissie Hsiao testified yesterday in the company’s antitrust trial that while ChatGPT has stolen market share from Google search queries, searches with “commercial intent”— like shopping—haven’t yet been significantly affected. But in notes from an October 2024 meeting presented in court, Google’s ad chief, Vidhya Srinivasan, was cited as saying that the “writing is on the wall” and Hsiao said Srinivasan believes it is “inevitable” that ChatGPT and other chatbots will eventually replace standard Google searches for many people.

I heartily agree. If you’ll indulge me, I’d like to share my own personal search story: People say women love to shop. I couldn’t relate less. I’ve barely set foot in a brick-and-mortar store for clothes, shoes, or accessories in nearly a decade. Instead, I’ve relied almost entirely on subscription boxes like Stitch Fix—and a lot of good old-fashioned Googling.

But that only gets me so far. Not only do I lack a passion for fashion, but I also typically feel like a styling failure. I’ve lost count of how many times I’ve dreamed of having Tan France from Queer Eye as my personal stylist. And I can’t even tell you how often I’ve fallen into a black hole of Google searches, desperately trying to figure out what goes with what—only to close the tabs in frustration.

Stuck at the search bar

I’m ready to buy—I’m just stuck at the search bar. Google, with its endless link scroll, is simply overwhelming for the shopping-averse. But it turns out that OpenAI’s ChatGPT, which debuted ChatGPT Search in October 2024, helps in ways that Google never could.

ChatGPT no longer relies on a static corpus of training data, with limited knowledge beyond 2023. Nowadays, ChatGPT offers real-time data courtesy of a blend of Microsoft’s Bing search engine (with which it has an integration) and OpenAI’s own web crawlers that scrape publicly available web pages (except those that block it via robots.txt) and update OpenAI’s internal knowledge base.

Here’s an example: Last week, I received a subscription box from Short Story, a Stitch Fix-like service for petites. When I couldn’t figure out how to style the six items I received, I turned to ChatGPT. Explaining my predicament, I got very specific, including my height, weight, and age. I did not include any photos of myself.

“Could I wear the black Souri blazer I got with the Brynn contrast border wrap skirt?” I asked.

Not only did ChatGPT reply in the affirmative and offer styling tips, it followed up by asking, “Would you like me to pull some top options that would work with this combo?” When I said yes, it served up links to several options from specific brands.

Without any prompting, it followed up by asking if I would like assistance in selecting accessories or footwear to complement the outfit. Would I? Only Tan France would be better!

The next day, I went back to ChatGPT to continue the conversation, turning to my recent struggle to camouflage my midlife stomach in an era of body-conscious tanks and tops. Again, not only did I get tips on everything from blouse shapes to shirt-tuck hacks, but I got a holy grail follow-up:

“Would you like me to send over a few real product links with visuals?” ChatGPT wrote. “I can pull some options under $75 or from brands you like (let me know if you have preferences like J.Crew, Quince, Madewell, etc).”

Thrilled, I threw a Hail Mary: “Do you have any in petite sizes?” Up came five more options. Inwardly cheering, I clicked through all the links, and there they were—actual, helpful items of clothing (yes, in petite sizes) that I would never have thought of or looked for. Success!

Building search Nirvana

So far it’s unclear whether brands are partnering with OpenAI to ensure their products are surfaced more prominently in ChatGPT search results, which would raise serious transparency and trust concerns—currently users expect ChatGPT to give them the best, most relevant answers, not ads dressed up as answers. My sense is not yet—although the Financial Times recently reported on OpenAI’s plans to ultimately introduce an advertising model. And, aside from explicit commercial relationships with OpenAI, many brands are trying to figure out if there’s a way to influence ChatGPT’s answers so that their products are more likely to get mentioned in outputs, in much the same way companies figured out ways to potentially game Google’s search algorithm through “SEO optimization.”

Apparently, OpenAI has big plans for building on the kind of search Nirvana I experienced, and it sounds like it’s even looking to Google to help it get there. On Tuesday, Nick Turley, OpenAI’s head of product, testified at the Google trial saying that improving OpenAI’s search offering is crucial for the company, which it says requires Google’s search technology, but that Google has declined to work with OpenAI (I can’t imagine why).

According to Bloomberg, during his testimony, Turley did not talk directly about Microsoft but referred to a search relationship with “Provider No. 1.” Turley said that OpenAI had “significant quality issues” with the search information from that provider—presumably Bing. Apparently in August, OpenAI reached out to Google about whether they could reach an agreement to gain access to the search giant’s index, Turley said, but it declined.

Perhaps Google wants to do what OpenAI is doing with its AI search, but so far, from my own experiments, it seems to be lagging. In trying Google’s AI mode, I had to ask specifically for product links, and there was no follow up to keep the conversation going or offer other options I might not have thought of. Perplexity, too, isn’t quite there yet either—in a similar search, it didn’t offer product photos or easy-to-read brand descriptions. It fussed quite a bit with unnecessary tables and bullet points, rather than ChatGPT’s friendly, personal stylist vibe.

Whatever the results of Google’s antitrust trial, and whether or not ChatGPT winds up being the big winner in AI shopping, it’s clear that we are entering a new era—and every brand is going to need to pay attention. Keep in mind, my experience wasn’t about an AI agent doing the purchasing for me. It was simply about discovering—in a natural, conversational way—what was always there, but I could never find in the deep blue sea of Google links. Hallelujah.

And with that, here’s the rest of the AI news.

Sharon Goldman
sharon.goldman@fortune.com
@sharongoldman

AI IN THE NEWS

Trump signs executive order on AI education. According to Bloomberg, President Donald Trump has signed an executive order aimed at ramping up artificial intelligence education across the U.S. Framing AI as essential to maintaining American dominance, the order creates a White House Task Force on AI Education, chaired by science advisor Michael Kratsios. The panel will include top officials from the Departments of Education and Labor, as well as AI and crypto adviser David Sacks. The order directs the task force to launch public-private partnerships focused on K–12 AI education and academic applications of AI tools. It also instructs the Education Department to prioritize AI in teacher training grants, encourages the National Science Foundation to fund AI-in-education research, and pushes the Labor Department to expand AI-related apprenticeships and certifications. High school students could soon see new AI coursework and credentialing opportunities as part of the initiative.

California Bar discloses AI was used to develop some questions in problem-plagued February exam. The State Bar of California revealed this week that a portion of its February 2025 bar exam included multiple-choice questions developed with the help of artificial intelligence—adding fuel to a growing controversy over the test’s disastrous rollout, the Associated Press reported. In a statement, the Bar said it will now ask the California Supreme Court to adjust scores for affected test-takers, many of whom were already reeling from widespread technical failures during the exam. Glitches ranged from platform crashes and frozen screens to lost essays and broken copy-paste functions. Out of 171 scored multiple-choice questions, 23 were created by ACS Ventures using AI tools—sparking outrage from legal educators. “Having the questions drafted by non-lawyers using artificial intelligence is just unbelievable,” said Mary Basick, assistant dean at UC Irvine Law School. “The debacle that was the February 2025 bar exam is worse than we imagined.”

Anthropic’s chief security officer says AI ‘employees’ are coming within a year. Jason Clinton, Anthropic’s chief security officer told Axios that AI-powered “virtual employees” could be active inside corporate networks within a year—a leap in autonomy that could create major new cybersecurity risks. While some CEOs already describe AI agents as digital employees that coexist with humans, Clinton described these AI identities as more advanced than today’s task-based agents. Unlike bots that respond to phishing alerts, virtual employees would have memory, defined roles, their own login credentials, and broad access across internal systems. That means this next-gen automation could transform enterprise productivity—but also open the door to massive vulnerabilities. “There are so many problems we haven’t solved yet from a security perspective,” Clinton cautioned.

Measuring the AI revolution. I like this chart-centric piece from the Wall Street Journal. While hundreds of millions of people worldwide use ChatGPT and tech companies are expected to spend $1 trillion over the next few years building out infrastructure to support AI services, it’s hard to visualize the technological and cultural transformation that happening. This provides four interesting charts that highlight the rapid revolution and who the key players are.

FORTUNE ON AI

Goldman Sachs execs’ images used in deepfake video campaign on Instagram —by Jim Edwards

Ex-OpenAI employees sign open letter to California AG: For-profit pivot poses ‘palpable threat’ to nonprofit mission —by Sharon Goldman

Exclusive: This Sequoia-backed AI startup can run thousands of voice interviews at once—and it’s raised $27M to disrupt market research —by Sharon Goldman

AI CALENDAR

April 24-28: International Conference on Learning Representations (ICLR), Singapore

May 6-7: Fortune Brainstorm AI London. Apply to attend here.

May 20-21: Google IO, Mountain View, Calif.

July 13-19: International Conference on Machine Learning (ICML), Vancouver

July 22-23: Fortune Brainstorm AI Singapore. Apply to attend here.

EYE ON AI NUMBERS

71%

That’s the number of AI tools used in the workplace that could put sensitive company data at risk, according to a new report from cybersecurity firm Cyberhaven. After analyzing over 700 tools, the company flagged the majority as high or critical risk.

What’s the danger? Many of these tools may accidentally expose employee interactions, training data, or even give third parties access to internal information—without strong safeguards in place. As AI adoption speeds up, companies will need to get serious about protecting their data.

I hate shopping. ChatGPT is already helping me in ways Google never did