AI news for Saturday, November 7, 2026: Anthropic's IPO price slides, deepfake CEOs take $129M

The market marked down the AI giants, fake CEOs on calls fooled a bank and a hospital system, and an AI handed over a proof you can rerun. Trust what you can check.

Business3 min
Anthropic IPO buyers talk $1.2T as roadshow nears
Potential buyers of Anthropic's IPO shares are discussing a $1.2 trillion to $1.3 trillion valuation, down from as much as $1.5 trillion, Bloomberg reports, and OpenAI has raised $11 billion at $1.25 trillion. The two labs you buy tokens from just got marked down. Expect them to chase revenue harder.
Why it matters A public Anthropic will answer to quarterly growth, and OpenAI is raising less than it asked for. If your margins depend on token prices falling every quarter, lock in the enterprise discounts both labs offer now and build a budget for the day prices stop falling.
SRCSEC EDGAR
Safety Policy3 min
Deepfake CEOs took $129M from companies, DOJ says
Federal prosecutors charged three people with using AI to mimic CEOs' voices and faces on calls, taking at least $129 million from companies between July 2024 and February 2026. If your finance team treats a video call as proof, that habit is now the hole.
Why it matters The fake CEO looked and sounded right, and hospital systems, a bank and a fund company still wired millions. Add one rule this week: every large or confidential payment request gets confirmed by calling back a number from your own directory, never one from the email or the invite.
SRCCyberScoop
Research Safety3 min
PSI says its AI proved a 2008 quantum conjecture
PSI says a research system running a non-public Anthropic model proved the Aaronson-Ambainis conjecture, open since 2008, and the Lean proof checker accepted it. If the mathematicians reading it agree, AI just settled a question researchers chased for 18 years, with a receipt anyone can rerun.
Why it matters A machine-checked proof is the one kind of AI output you do not have to take on faith. If you review AI-generated work, code or math, this is the standard to ask for: a check you can rerun yourself, published next to the claim.
SRCPSI
Policy Hardware3 min
House chairs ask Nvidia's Jensen Huang for spy briefing
Three House committee chairmen asked Nvidia CEO Jensen Huang to brief them behind closed doors on Nov. 18 about his dealings with a Megaspeed executive they call a reported Chinese intelligence asset. A newspaper allegation is now a congressional file, with a deadline.
Why it matters If your roadmap depends on Nvidia hardware, the chip you rent does not change this month. The rules around who can buy it may, and this request puts Nvidia's sales to regional clouds in front of three committee chairmen. Plan as if those rules get stricter.
SRCHouse Foreign Affairs Committee
Models Products3 min
Meta says new Muse models forget 2.4x less
Meta is rolling out versions of its models trained with Muse Wonder to Muse and Meta AI, and says they score 47% on its own index, up from 38%. Meta finally published numbers. It also wrote the tests.
Why it matters If you use Muse or Meta AI for planning and errands, its memory is supposed to get better soon. Test it on facts you can check before you lean on it, because no outside benchmark backs these numbers yet.
SRCMeta AI Research

ShareX LinkedIn Hacker News Email

Prompt of the day

Write a payment rule a deepfake CEO can't talk past

› Federal prosecutors say a group used AI to fake CEOs' voices and faces on calls and talked companies into wiring at least $129 million. I want a payment approval rule my team can follow so that can't happen to us.

About us: [company size], [who can approve payments today], [tools we use for payments and chat, e.g. bank portal, Slack, email], [typical wire sizes].

Please write:
1. A one-page payment verification policy in plain language: which requests trigger extra checks (amount threshold, new payee, changed bank details, any request called urgent or confidential), and exactly how to verify (call back on a number from our own directory, never one from the email or invite, second approver, waiting period).
2. A short script an employee can say out loud to a "CEO" on a video call to pause a request without feeling rude.
3. Five red flags from this kind of scam (fake e-signature emails, an NDA before a payment, a handoff to an outside "lawyer") written as a checklist.
4. A 10-minute training exercise to run at our next team meeting.

Keep it short enough that people will actually read it. Ask me questions first if anything about our setup is unclear.

Open in Magai

Works in any AI chat. In Magai, run it against several models at once.

Get every edition

This one arrived in inboxes at 6am ET.

Anthropic IPO buyers talk $1.2T as roadshow nears

Deepfake CEOs took $129M from companies, DOJ says

PSI says its AI proved a 2008 quantum conjecture

House chairs ask Nvidia's Jensen Huang for spy briefing

Meta says new Muse models forget 2.4x less