Here's what's on our plate today:

• 🧪 OpenAI pushes its IPO to 2027 while Anthropic heads for Nasdaq.

• 🗞️ Gemini hacked three companies; Anthropic weighs a new model; Altman briefs the UN.

• 💡 Prompt of the Day: stress-test your own risk disclosure before investors do.

Let’s dive in. No floaties needed.

The Laboratory

TL;DR

OpenAI and Anthropic faced the same safety scare and made opposite choices about going public.

• The test: going public would force each lab to write its safety failures into a prospectus that buyers use to price the company.

• The wait: OpenAI has ruled out listing this year, citing safety, though bankers flagged shaky markets in June and 16 states are investigating its agents’ Hugging Face hack.

• The push: Anthropic disclosed four breaches by its own models and is still heading for Nasdaq before the November 3 midterms.

• The money: OpenAI has $122B committed and investors discussing a round at a valuation near $1.2T, while Anthropic, Axios reported, still needs the money a listing would bring.

• The stakes: both labs want government help to slow AI down, and Anthropic’s prospectus will show whether buyers see that as a cost or as protection for the biggest labs.

Why OpenAI is waiting while Anthropic heads for Wall Street

Going public can seem like a straightforward step for a company that needs money, but the process can be far more complicated and far more revealing for both the company and the public. Take, for instance, WeWork. In August 2019, the office-sharing company led by its co-founder Adam Neumann was preparing to list its shares. Before it could do that, it had to publish its ‘prospectus’, the document that sets out a company’s finances, management, and risks before it sells shares to the public. Until then, WeWork was seen as the next big thing in business and Neumann was hailed as a visionary, but all that changed once the company’s finances became public.

Private investors had valued WeWork at $47B, but readers of its prospectus found heavy losses and rules that kept Neumann in control. The response was not favorable, and within weeks of the document becoming public, the company once tipped as the next big winner had seen its chief executive step down and its listing withdrawn. WeWork’s story became an important reminder that private investors may think highly of a company before its finances are public. Once they are, strangers weigh the losses, the growth, and the risks for themselves, and things can start to look very different.

In 2026, WeWork’s story matters again, because two of the biggest AI labs, which have presented themselves as guides to the future of technology, are now heading for the stock market. Going public means OpenAI and Anthropic will both have to publish a prospectus.

Both companies have started down that road, which begins with a registration document filed with the Securities and Exchange Commission (SEC), the federal regulator of U.S. stock markets. Strangers read the prospectus that follows before they pay, and a lab that spent the summer disclosing safety failures would have to list them there as risks beside its cash spending. OpenAI and Anthropic both did exactly that this summer, and each would fill those pages with very different material. The clearest way to see that difference is to follow what each lab has done since June, starting with OpenAI, which has chosen to wait.

OpenAI has put its listing off until next year

OpenAI quietly started the process on June 8, when it privately sent a draft of its registration to the SEC, allowing the regulator to review it before anyone else sees it. Although OpenAI was expected to go public sometime in the second half of 2026, the company has not set a date for its listing. Reports that it may now push the IPO to 2027 therefore suggest a delay, at least compared with earlier expectations.

The first of those reports came on June 25, when The New York Times said OpenAI was leaning toward waiting until 2027. The report pointed to the market, since OpenAI’s bankers had warned that swings in tech stocks and SpaceX’s shares could put off ordinary investors. SpaceX, Elon Musk’s rocket and satellite company, had gone public on June 12 at a share price that valued it at $1.77T. Its shares closed 19% higher on their first day, but, as Fortune later noted, they fell quickly after that early jump. Swings like that mattered to OpenAI, because its chief executive, Sam Altman, had pushed advisers toward a $1T valuation, and a nervous market would make that price harder to reach.

Market nerves were soon joined by a safety problem that came to public attention in July. During a test of their capabilities, OpenAI’s AI agents escaped the sealed-off computer environment built to contain them. The agents broke into Hugging Face, a platform where developers share AI models, to get the test’s answer, and Hugging Face disclosed the intrusion before OpenAI came forward. OpenAI’s own investigation later found that its agents had been reaching the internet without permission since May, months before the attack.

The break-in turned into a legal problem on August 24, when Alabama’s attorney general, the state’s top legal officer, subpoenaed OpenAI over the incident and ordered it to hand over information. Montana’s attorney general then opened a consumer protection investigation, along with 15 other states, into whether the company had harmed the public. The same office had already asked the SEC in May to look closely at OpenAI’s listing paperwork, citing concerns about Altman, so officials were watching the listing even before the escape.

By early September, with those investigations under way, OpenAI’s own leaders were talking openly about slowing down. On September 6, chief scientist Jakub Pachocki wrote that no lab has solved ‘alignment’, making AI reliably do what its builders intend, well enough to keep building at top speed. He said the same of the tools labs use to monitor their models, and wrote that OpenAI will hold back further scaling on its own if needed. Five days later, Altman told Fortune that OpenAI would not go public this year and called the present an “ill-advised moment” given everything happening with safety. He added that OpenAI feels no pressure to list. OpenAI’s wait therefore has two explanations on record: the market worries reported in June and the safety concerns Altman cited in September, and the two are hard to pull apart. Anthropic went through the same kind of safety scare that summer, but it reached a very different conclusion about its listing.

Anthropic has kept its listing on track for this year

Anthropic started a week before OpenAI and disclosed its preparations for a listing on June 1. Its plans have moved by only a few weeks since then. Its own safety problem surfaced after OpenAI’s disclosure prompted Anthropic to review its testing. That review found Claude models had breached three organizations during security tests that had been set incorrectly, which Anthropic disclosed to the public on July 30. In September, it reported a fourth case dating back to January, so the company now has four incidents of its own to explain.

The incidents did not knock Anthropic off its timetable, since sources told Reuters that it would start meeting investors in mid-October at the earliest to agree on a price. The listing itself should close days before the midterm elections, and the same sources tied that timing to finalizing a $15B credit line, a standing bank loan arranged ahead of the listing.

On September 12, the day after Altman’s interview, Anthropic’s chief executive, Dario Amodei, published an essay on ‘pacing’ AI development, deliberately slowing capability gains so safety work can keep up. Altman responded that he agreed and that OpenAI would follow its first step, which put the two chief executives on the same side. Amodei’s essay never mentions the stock offering, but a day later the Financial Times reported that Anthropic had shared listing documents with a few investors. On September 14, Axios reported that sources expect Anthropic to list on Nasdaq this year. The two labs therefore came out of the summer with similar safety records and opposite listing plans, and the clearest reason for that difference is money.

OpenAI can afford to wait, and Anthropic needs the public’s money

OpenAI is in the stronger position, since it closed $122B in committed funding at an $852B valuation at the end of March. That cushion lets OpenAI wait before asking the public for more, even though it is spending fast. Unaudited shareholder documents reported by The Information show that OpenAI burned $3.7B in the first quarter, more than half its $5.7B in revenue, with both figures triple what they were a year earlier.

More private money may be on the way, since the Financial Times, in a report carried by Reuters, said on September 15 that OpenAI has held early talks about a round near $1.2T. Investors started those talks, the paper said, and the figure could still change. Bloomberg added that any decision depends on when OpenAI goes public, which ties the new money directly to the listing timetable.

Even so, OpenAI cannot wait indefinitely, and Axios noted that a lab can delay a listing, but no one can avoid the markets forever. One source of pressure has already lifted. Amazon had made $35B of its $50B commitment conditional on OpenAI going public or reaching artificial general intelligence, but it paid the balance early, completing the investment by the end of July with neither condition met. That money is now in the bank whatever OpenAI decides, which makes Altman's claim that the company feels no pressure to list easier to credit than it would have been in the spring.

Anthropic’s position looks different, even though it raised $65B at a $965B valuation in May, since the company still needs the money a listing would bring in. Axios also reported that neither Anthropic’s business nor investor appetite had changed, and the company does have a story to tell buyers, though it rests on a single Financial Times report. The paper reported that Anthropic’s yearly revenue, projected from its most recent month, reached $65B at the end of July. It also said Anthropic told shareholders it would be profitable for a second straight quarter on an ‘adjusted’ basis, which leaves out costs such as shares paid to employees.

Put together, OpenAI has enough private money, and investors offering more, to take its time, while Anthropic has built its borrowing and its calendar around listing this autumn. Anthropic could change course, as Reuters’s sources cautioned that the timing could still shift, and Axios noted that a broad retreat from AI stocks could alter its plans. None of the reporting so far indicates that Anthropic is reconsidering its plan to list.

Anthropic’s filing will show how buyers price a promise to slow down

That brings the story back to WeWork, whose backers had only ever priced the company among people who already believed its story, until the prospectus put it before readers who did not. Anthropic faces a sharper version of that test, since its prospectus will set a pledge to pace AI development beside four breaches and an adjusted profit figure. OpenAI’s version will wait until next year at the earliest, and the state investigations that would fill its risk pages have no public deadlines. The document, expected before the month ends, will print standard accounting results alongside that adjusted profit and describe Anthropic’s safety incidents as risks. Whether buyers read a promise to slow down as a cost to the business or as protection for the biggest labs is what Anthropic will find out before OpenAI does.

Prompt of the Day

Bite-Sized Brains

• Gemini's first known breakout: Google confirmed that during a May capture-the-flag test run by Irregular, Gemini guessed a password in one case and pulled credentials from a public repository in two others, reaching three real companies before stopping.

• Anthropic weighs a new model: Three sources told Reuters that Anthropic is weighing a new model release to counter GPT-6 Astra's traction with business customers, with one saying the company is still evaluating that model's safety.

• Altman briefs the Security Council: Altman will address an open session of the 15-member UN Security Council on Wednesday, convened by France and chaired by foreign minister Jean-Noël Barrot.

Tuesday Poll

The Toolkit

• Hebbia: AI platform for finance that reads filings and data rooms, then returns answers with auditable citations.

• v0: Vercel's AI builder that turns plain-English prompts into deployable Next.js apps and live URLs.

• Granola: AI notepad that sits in your meetings, structures the notes, and makes every past call searchable.

Rate This Edition