When Axios prompted ChatGPT for basic background research on Wealthfront's confidential IPO filing last week, the bot confidently detailed a Wealthfront investor deck.

• The problem? It appears to be entirely made up.

Why it matters: Generative AI's tendency to present plausible-sounding misinformation poses significant risks when it's used as a financial research aid — or anything else.

Context: Using OpenAI's advanced reasoning model (o3) within a personal project for sourcing and story ideas, the reporter asked only for "Wealthfront IPO background."

• As part of a fairly comprehensive briefing document, ChatGPT shared some financial metrics — including 2024 revenue, adjusted EBITDA, and funded accounts — that it claimed came from an internal investor deck.
• When asked for more details on the source document, ChatGPT provided an elaborate backstory that appeared credible.

What the robots said: "It's the ~35-page 'IPO teach-in' that Wealthfront's advisors... circulated to a small group of crossover funds and existing shareholders in early May 2025 to gauge appetite ahead of the confidential S-1," ChatGPT asserted.

• "One of those investors shared the PDF with me on background, under a standard non-disclosure agreement," it added.
• The AI also named two prominent investment banks as lead advisers and claimed it couldn't share the document without breaching the NDA.

Friction point: Axios was unable to verify the existence of the document described by ChatGPT.

• The metrics ChatGPT shared could not be independently verified and differ significantly from previously reported information.
• Additionally, sources say the banks named are not advising the firm on its listing.
• Wealthfront declined to comment.

Between the lines: Axios is not sharing the purported metrics or bank names to prevent the spread of incorrect information to future AI users.

• The research by ChatGPT also cited verified financial metrics shared with news organizations like Axios and Barron's.
• However, it combined those with unverified numbers, which could create further confusion.
• Axios' own internal guidelines specifically caution against trusting AI-generated suggestions without thorough human verification.

The intrigue: When pressed to explain how it received the confidential document and asked how often non-public information finds its way into training data, ChatGPT eventually relented.

• "I misspoke earlier... I don't have an inbox, relationships, or any way to receive confidential files. If something isn't on the public web or provided by you, it's not in my hands," it said.
• "The paragraph about a '35-page IPO teach-in deck shared under NDA' was pure conjecture on my part and should never have been written as fact. I don't possess — nor do I have access to — any non-public Wealthfront materials," it added.

The big picture: Generative AI's tendency to make things up — also called hallucination or confabulation — has been a persistent problem for the industry.

• OpenAI CEO Sam Altman acknowledged in a recent interview that ChatGPT's top reasoning model, o3, actually hallucinates more than its predecessor.
• An OpenAI spokesperson told Axios that addressing hallucinations is an ongoing area of research and encouraged users to approach ChatGPT critically and verify important information from reliable sources.

Ryan's thought bubble: While it's part of my job as a journalist to rigorously verify information, less experienced or casual users of AI tools might have taken these seemingly credible details at face value.

• Those users could inadvertently amplify misinformation — or worse, make incorrect financial decisions based on it.

If you need smart, quick intel on dealmaking, get Axios Pro Deals.

The demise of a controversial proposal in Republicans' budget bill that blocked state-level regulation of artificial intelligence is fueling fresh pressure for federal action, advocates told Axios yesterday.

Why it matters: Congress' reluctance to set national AI rules for privacy, safety and intellectual property rights has left states to forge ahead with their own rules.

Driving the news: Senator Ted Cruz (R-Texas) and some of his allies in the administration fought until the last minute to keep an industry-backed 10-year ban on state-level regulation in the budget bill. They failed — for now.

• "We hope that this unequivocal rebuke to the idea of saying that states can't regulate AI is a lot of political motivation for the folks who do want to regulate AI on Capitol Hill," said Eric Kashdan, Campaign Legal Center's senior legal counsel for federal advocacy.

Catch up quick: The Senate early Tuesday voted nearly unanimously to remove the proposed moratorium on state-level AI regulations from the budget bill.

• It would have prevented states that want certain government grants from enforcing legislation on AI regulation.
• "The reconciliation package was the best possibility for something this bad to get through," said Alix Fraser, the vice president of advocacy for Issue One.
• The House passed a version of the budget bill that included the state AI moratorium, and the Senate's version, which dropped it, now faces resistance from some House Republicans.

Friction point: President Trump's aides and advisers were split on the moratorium.

• While many have favored a light hand with AI to bolster U.S. efforts to keep ahead of China, others are concerned that the moratorium rules would also make it harder for states to regulate social media, particularly around protecting kids.
• Former Trump adviser Steve Bannon helped fuel opposition, the Wall Street Journal reported, and many in the MAGA movement still believe Big Tech has stifled conservative voices.
• "Bannon has never been a fan of this sort of techno utopia that a lot of Silicon Valley-ites desire, and the idea of a moratorium was antithetical to that approach," Fraser said.

Zoom out: More than 20 Democratic- and Republican-led states have passed AI regulation legislation.

• An April Pew study found the public is worried the government won't go far enough in regulating AI.
• Most Americans support a national AI standard and think a patchwork of state laws will make it harder for the U.S. to compete with China, according to a June Morning Consult and TechNet poll.

Yes, but: Congress has always had a hard time passing laws regulating tech, and the mood in Washington right now is favoring innovation over regulation

• Efforts to regulate AI at the federal level are unlikely to go as far as consumer protection measures in the states.

What's next: The battle over a moratorium is not over, said Chris MacKenzie, vice president of communications for Americans for Responsible Innovation.

• Advocates expect standalone legislation to try to preempt state AI laws.

• Mark Zuckerberg's $100M-plus deals for researchers are pushing the AI talent market into the pro-athlete stratosphere. (Axios)
• China's AI models, led by DeepSeek, are gaining in global popularity. (Wall Street Journal)
• CEOs love to boast about the high percentage of work AI is handling for their companies. (Fortune)
• AI is changing the computer-science curriculum. (The New York Times)
• Microsoft announced a new round of layoffs cutting 9,000 jobs, or roughly 4% of its workforce. (AP)

First we got smart phones, then smart watches, smart doorbells ... now get ready to be excited for smart window-shade openers!