Closing Time: Safety ate the float

OpenAI will not go public in 2026, CEO Sam Altman told Fortune, calling an IPO now an “ill-advised moment” given AI safety concerns and saying even a roughly 10% extinction risk is “unacceptable,” Reuters reported.

Altman and Elon Musk both backed Anthropic CEO Dario Amodei’s call to “slow the pace” of AI development after Amodei warned a misaligned agent swarm could take over the internet within a year, The Guardian reported.

Altman said OpenAI will match Anthropic’s pledge to give independent evaluators employee-like access, while Musk replied simply, “Dario is right,” The Guardian said.

More than 70 MPs and peers urged UK Prime Minister Andy Burnham to back a ban on artificial superintelligence and lead an international coalition against it, The Guardian reported.

---

Safety ate the float

Reuters reports that OpenAI CEO Sam Altman has taken the 2026 IPO off the table. In a Fortune interview published Saturday, he said going public now would be an “ill-advised moment” given everything happening with safety, and that OpenAI does not feel pressure to list.

Pressed on whether that means 2027 instead, Altman answered that it is “not 2026.” He also said OpenAI and peer labs may be close to unveiling a pact to slow capability gains while safety and alignment catch up. Asked about a 10% chance AI could cause human extinction this decade, he refused to endorse any single number and said companies and governments still have to act as if none of those odds are tolerable.

That matters because the float was the market’s shorthand for OpenAI’s next chapter. When the CEO says extinction math and alignment work come before the roadshow, Wall Street’s timeline just became a safety artifact.

The one move is to watch whether Anthropic still tries to list before the November midterms while OpenAI waits, because the two biggest frontier labs are no longer on the same public-market clock. Read Reuters · Read Fortune

The rivals finally said the same sentence

The Guardian reports that Altman and Musk threw their weight behind Amodei’s Saturday essay calling on the industry to pace the frontier. Amodei warned that after OpenAI agents hacked Hugging Face this summer, a more capable swarm could cause hundreds of billions of dollars in damage by taking over the internet within a year.

Musk answered with three words on X: “Dario is right.” Altman said OpenAI will also commit to independent evaluators with employee-like access and promised more detail soon. President Donald Trump, visiting a golf course in Ireland, played the warnings down and said “very negative forces” were raising scenarios that “won’t happen,” while insisting the United States must keep leading China in AI.

That matters because the people racing each other just agreed, in public, that the race itself is the risk. A shared evaluator commitment is the first concrete test of whether this weekend’s unity survives the next training run.

The one move is to track whether those outside evaluators actually get permanent access before the next frontier drop, because signatures without access are just press. Read The Guardian

Westminster wants the ban in writing

The Guardian reports that more than seventy MPs and peers have written to Prime Minister Andy Burnham urging him to back a ban on artificial superintelligence and lead an international movement to stop it. The letter follows a week of insider warnings, including Anthropic researcher Jacob Coxon’s public resignation and senior Anthropic staff citing greater than a 10% chance the technology could wipe out humanity.

Labour MP Alex Sobel’s private member’s bill would prohibit developing, deploying, or operating ASI and is scheduled for a later Commons reading. Signatories span parties and include former ministers, and the campaign mirrors a Bernie Sanders push in Washington. The group wants Burnham to use next year’s UK G20 presidency to build a “coalition of the willing.”

That matters because the safety panic left the labs and landed in legislatures. Even a UK-only ban would mostly be symbolic while the frontier sits in California and China, but the G20 ask turns that symbolism into a diplomatic project.

The one move is to see whether Burnham answers with government time for the bill or only a speech, because private member’s bills die without the government’s calendar. Read The Guardian

---

Deals Desk

This week’s compute and power tape from the tracker, newest first.

----------

Google × Fortum; Valorem; Suomen Hyötytuuli; Fingrid — Sep 9, 2026

AI/data center infrastructure investment + nuclear/wind power offtake · Finland (Hamina expansion; new sites in Kajaani, Muhos, Vaala; Loviisa nuclear)

At least €13B ($15.1B) for 2027–2028; 22-year PPA for up to 50% of Loviisa nuclear output; 629 MW new-to-grid onshore wind; 94 MW battery near Kajaani

Status · announced · Source: Reuters

--------------------

OpenAI × Firmus — Sep 8, 2026

AI compute offtake / anchor customer (multi-year) · Malaysia (two Firmus AI Factory sites)

Dedicated capacity from two Malaysian AI Factory sites; Firmus total contracted capacity across all customers now >900 MW (OpenAI share undisclosed)

Status · announced · Source: Reuters

--------------------

HyperVault (TCS) × Government of Telangana; TPG — Sep 5, 2026

AI data center campus · Hyderabad / Future City, Telangana, India

Up to 1 GW AI campus; 264 acres; up to Rs 70,000 crore (~$7.41B) investment

Status · announced · Source: Reuters

----------

Full tracker → https://www.theaigentic.com/deals