1Anthropic's IPO numbers reach the banks - and are withdrawn within hours
Anthropic circulated an IPO filing to investment banks on Monday outlining plans to sell roughly 867 million shares at $70 to $80 each, raising as much as $69.3 billion in what would rank among the largest public offerings ever, Reuters reports. Hours later the company withdrew the document, telling banks the pricing information was inaccurate and that revised details would follow. The IPO itself, with a roadshow, is still targeted for next week.
For context: the figures come from a document Anthropic itself says was inaccurate, so treat $70 to $80 as a draft, not a price. Anthropic's updated S-1 earlier this month showed revenue of about $33 billion in the first nine months of the year.
In plain English
Anthropic's first concrete price tag escaped into the open - and the company pulled it back the same day.
Why it matters
This is the first hard number in the test of whether public investors will pay up for AI at scale. Last week the question was whether Anthropic could hold a $1.5 trillion valuation after the AI stock selloff; a hasty same-day retraction suggests even the company hasn't settled what the market will bear.
SourcesReuters' exclusive·Business Standard's report
2Eight countries and the big AI labs sign a Paris safety pact
Representatives of eight countries signed the "Paris Call for AI Safety and Security" on Monday, alongside OpenAI's Sam Altman, Anthropic's Dario Amodei, Microsoft's Brad Smith, Mistral's Arthur Mensch and AI pioneer Yoshua Bengio; US Treasury Secretary Scott Bessent also joined, and Germany did not sign. Signatories commit to shared evaluation procedures for powerful models, standardized publication of non-sensitive test results, disclosure of known risks, support for independent audits, and shutdown mechanisms for models under their control.
In plain English
Governments and the labs agreed, on paper, that the strongest AI systems need brakes and outside inspectors - with no deadlines and no penalties.
Why it matters
The pact is voluntary, but it is the first time the major US labs have jointly signed up to shutdown mechanisms and outside audits at a government level, and a US cabinet secretary signing marks a shift from Washington's earlier rejection of international oversight. The Paris Peace Forum takes up AI regulation on Tuesday.
SourcesThe Elysee's declaration text·The Decoder's report
3Meta ships Muse Ember, a small model built for the work agents repeat all day
Meta's Superintelligence Lab released Muse Ember on Monday, its lightest and cheapest model yet, priced at 10 cents per million input tokens and aimed at the repetitive work agents do: tool calls, coding subtasks, routing, extraction, and long-document parsing. It handles text, images, audio and video, and Meta positions it as the volume layer under its frontier Muse Wonder - "Muse Wonder plans, Muse Ember handles the volume." It is the fourth model in the Muse family.
In plain English
Meta is betting that most agent work doesn't need a frontier brain, just a cheap, reliable pair of hands.
Why it matters
Cost per task is becoming the competitive axis for agents; Meta's own benchmarks put Ember ahead of Anthropic's Haiku and OpenAI's Luna per dollar, though Anthropic's model still wins on Meta's real-coding measure. Vendor benchmarks aside, a big lab pricing a capable multimodal model this low pressures everyone's agent economics.
SourcesTechCrunch's report
Checked against original announcements and named press reports on the morning of November 10, 2026. The Anthropic figures come from a filing the company circulated to banks and then withdrew as inaccurate; revised details have not yet been published, and the offering has not priced. The Paris declaration is a political self-commitment with no deadlines or sanctions. Meta's cost and benchmark claims for Muse Ember are the company's own measurements.
