In partnership with
Hey,
Here's the pattern I keep seeing this week: a huge claim lands, the internet loses its mind for a few hours, and then someone who actually understands the details starts asking uncomfortable questions.
That happened twice. OpenAI said its AI cracked one of math's seven $1-million problems until the mathematician whose unpublished research it may have leaned on called foul. And while everyone argued about that, Apple quietly shipped a new Siri built partly on a rival's model, in every language except the one used by roughly 450 million EU citizens.
Here's what's actually confirmed, what's still contested, and what it means if you build with this stuff.
In this issue:
Everything GTM. One platform.
Small teams don't have time to stitch together five tools and hope it works.
Apollo gives you everything you need to find leads, reach them, and close deals — all in one place:
230M+ verified contacts
AI-powered outreach
Data enrichment
Inbound lead capture
Meeting scheduler
And more
Stop juggling tools and start building pipeline that scales.
With Apollo, the AI revenue engine powering 4M+ users.
In this issue TODAY CONTENTS :
OpenAI's math "proof" — and the credit fight nobody's settled yet
Apple's first foldable, and the more interesting story hiding inside it
Google ties its AI ambitions to a Finnish nuclear plant, through 2050
Mistral raises Europe's largest-ever tech funding round
Quick hits: DeepSeek, XPeng's robot, and malware draining AI accounts
🏦 THE BIG STORY
OpenAI says it solved a $1M math problem. The mathematicians involved disagree.
On Tuesday, OpenAI announced that an unreleased model — one it says is meaningfully more capable than GPT-6 Astra — produced a proof related to the Navier-Stokes equations, one of the seven Millennium Prize Problems the Clay Mathematics Institute has offered $1 million each for since 2000. Only one has ever been solved.
The number that makes this newsworthy on its own: OpenAI says the proof came from roughly 10,000 AI agents working in parallel over 88 hours, at a compute cost north of $22 million. The result was checked in Lean, a formal proof-verification language, and published alongside a 166-page manuscript on September 8th.
Here's the asterisk that matters: what the model actually proved is a variant of the problem — a singularity forming in finite time under a smooth, deliberately-added forcing term, not the unforced version the Clay Institute is actually offering money for. OpenAI isn't claiming the $1 million, and independent mathematicians haven't had time to verify the work.
Then it got messier, and this detail matters: Buckmaster's collaborator is Levent Apogee , a mathematician affiliated with Anthropic — one of OpenAI's biggest competitors. The two say they'd spent months chipping away at the same underlying approach using AI tools from both companies, and were closing in on a result, when OpenAI's team caught wind of their progress around September 1st and moved quickly to finish its own version using, in Buckmaster's account, a similar line of attack. OpenAI has denied that its model was trained on or drew from the pair's private, unpublished work.
Nobody outside the two parties can settle that yet. But the bigger question is worth tracking regardless of how this specific dispute resolves: when a frontier lab's own tool is what researchers use to do unpublished work, what happens when the lab notices what you're working on? Fields medalist Terence Tao has spent much of this year warning about a related problem — AI systems that can generate results faster than the field can verify or even read them. This is that warning, playing out with $22 million and a public credit dispute attached.
Where this leaves things: treat "AI solves a Millennium Prize problem" headlines — including this one — as unverified until independent mathematicians weigh in. That's not cynicism. It's just the current, actual status.
AI NEWS HIGHLIGHT
DeepSeek says its new Flash model beats its own flagship. Nobody's independently checked that yet. DeepSeek opened a 48-hour public test (Sept 8–10) of a build called V4.1 Flash to anyone with an API key — no waitlist, just a model-name swap. DeepSeek's own release notes claim it now outperforms the pricier V4 Pro on performance, cost, and speed combined, and the company says it'll retire V4 Pro as a result. Worth being precise about what that is: a vendor's claim about its own model, with no independent benchmarks yet published. It's also natively multimodal, where the current Flash line is text-only.
XPeng's humanoid robot has a production line now. IRON — the robot that went viral last year after skeptics assumed a person was inside the suit — walked off a newly commissioned, 80%-automated assembly line in Guangzhou under its own power. One finished unit isn't mass production, but XPeng says volume output is still targeted for year-end, with commercial deliveries starting in 2027.
Anthropic is warning Claude users about session-hijacking malware. Commodity infostealers — Vidar, LummaC2, StealC, RedLine, and Acreed on Windows; Atomic Stealer on a small number of Macs — are lifting authenticated browser session cookies from infected computers and using them to burn through paid Claude usage, bypassing two-factor authentication entirely since the stolen token is already logged in. Anthropic says its own systems weren't breached — this is happening on infected personal devices. If your usage has looked off lately, it's worth scanning for malware and re-authenticating from a clean machine.
Also worth knowing: DermaSensor, an FDA-cleared handheld scanner that reads light reflected off a skin lesion, is getting fresh attention as a real-world example of point-of-care AI screening. To be clear about what it actually is: it's a tool for a physician to use on a lesion already flagged as suspicious, to help decide whether it needs a specialist look — not a diagnosis, and not meant for self-screening.
TECH HOT UPDATE
Apple's first foldable is here. The more interesting story is what's running inside it.
Apple's new CEO, John Ternus — who took over after Tim Cook's 15-year run ended earlier this month — used his first major keynote to unveil the iPhone Duo, Apple's first foldable and reportedly its biggest physical redesign in two decades. It opens to a 7.6-inch inner display, folds to 5.4 inches, and starts at $1,999 — about $700–800 more than the also-new iPhone 18 Pro. Pre-orders open October 16th; it ships October 23rd.
The fold is the headline. Siri is the story for this audience.
Apple is shipping a rebuilt "Siri AI" as a free iOS 27 update on September 14th, available on any iPhone that supports Apple Intelligence — not just the Duo — with access to your contacts, calendar, messages, email, and other apps so it can act across them instead of just answering questions. Processing happens on-device where possible and through Apple's Private Cloud Compute otherwise.
Three details worth knowing before you write your own take:
Multiple reports point to a custom Google Gemini model handling a meaningful share of the work behind the new Siri, alongside Apple's own on-device models. Apple hasn't confirmed this directly.
Siri AI ships as an English-only beta. French, Japanese, Korean, Portuguese, and Spanish arrive in October.
It's not available in the EU at launch, on any device — a detail Apple stated plainly in its fine print and one that's gotten surprisingly little coverage.
There's also a pricing wrinkle worth flagging: some of the newest AI camera features are exclusive to the cheaper iPhone 18 Pro, not the $1,999 Duo. If you're doing a "which iPhone should you actually buy" piece, that contradiction is your hook.
The Slop-Proof Audit
Secure your AI accounts — before the next infostealer finds them
Prompt: Anthropic just disclosed that infostealer malware — Vidar, LummaC2, StealC, RedLine, Acreed — stole active Claude session tokens from infected user devices and consumed usage limits. The attack succeeded not because Anthropic's infrastructure was breached, but because browser-stored session tokens on infected PCs gave attackers authenticated access. Claude Sonnet 5 is now $3/$15 — stolen AI credits have real monetary value. This is the same pattern as CoSnitch (Copilot memory poisoning) and the DPRK npm campaign. The AI credential attack surface is real and growing.
Our team uses the following AI tools with stored credentials or sessions: [list every AI service — Claude, ChatGPT, Cursor, GitHub Copilot, Gemini, Perplexity, and any API keys stored in config files or environment variables].
Help me secure our AI credential surface across three areas:
1. The credential audit — for every AI service we use: where are the credentials stored? Browser session tokens, saved passwords, API keys in .env files, config files in repos, or hardcoded strings? For each storage location: what is the blast radius if that location is compromised by commodity infostealer malware? The five malware families in today's Anthropic incident target exactly these locations.
2. The hardware key rollout — OpenAI made hardware security keys mandatory for all Daybreak accounts on September 1. Anthropic supports FIDO2 hardware keys for Claude accounts. GitHub supports them for Copilot. For our team: which AI accounts should require hardware keys immediately, and what is the rollout order based on credential value and blast radius? Write the five-step rollout plan.
3. The API key hygiene — API keys in .env files, hardcoded in scripts, or committed to repositories are the highest-risk AI credentials. For our codebase: what is the audit process to find exposed AI API keys, what rotation schedule should we set, and what monitoring would alert us if an API key is being used from an unexpected IP or at an unusual volume? Stolen AI API keys can consume thousands of dollars of credits before the bill arrives.
End with a one-paragraph security posture statement for our AI stack — what we have secured, what remains at risk, and the single action that reduces our exposure the most for the least effort. If the answer is "enable hardware keys on our primary Claude and OpenAI accounts," say so directly.WHAT’S HAPPENING AI TODAY
Google just tied its AI ambitions to a nuclear plant, through 2050
Google announced a €13 billion (~$15.1B) investment in Finnish AI and digital infrastructure through 2027–2028 — its largest single investment in Europe to date. The actual mechanism: a 22-year power purchase agreement with Finnish utility Fortum for up to 50% of the output of the Loviisa nuclear plant.
That deal is the real story. Without it, Fortum says Loviisa was headed for shutdown around 2030. With it, the plant runs through 2050, and Fortum gets roughly €1 billion in committed revenue for a modernization program. Google's President and Chief Investment Officer, Ruth Porat, has taken to calling the approach bring your own power — and confirmed it's Google's first nuclear deal outside the US.
The investment covers four data center sites (Hamina, Kajaani, Muhos, and Vaala), plus 629 MW of new wind capacity, battery storage, and a €31 million community fund. Google says it'll add roughly €3.6 billion to Finland's GDP during construction and support about 7,000 jobs a year once running.
Not everyone's celebrating. Finnish opposition politicians are warning the deal could strain the national grid and push up electricity prices, and are pushing for national permitting rules to manage data center growth going forward. That tension — one company's compute demand now large enough to shape national energy policy — is going to resurface elsewhere. Finland is just the first place it's this visible; Microsoft and ByteDance are both reportedly eyeing the same low-carbon power.
Exploring AI Voice With SuperBloom
For Deel's "Feeling of Deeling" campaign, agency SuperBloom needed one consistent brand voice across markets—deployed fast, without sacrificing quality or consent. In this on-demand video session, SuperBloom and Voices break down the casting, production, and governance, plus what they'd do differently and where AI voice is headed next.
🧰 TOOLBOX
NotebookLM — Load your research and sources; get grounded answers and audio overviews you can listen to on a walk.
Descript — Edit video and podcasts by editing the transcript. Deleting a bad take feels like deleting a sentence.
Opus Clip — Turns one long video into a week of short-form clips, with auto-captions and reframing.
Suno — Custom music from a text prompt, for intros and background beds.
Gamma — Outline in, polished deck or one-pager out. Fastest path from idea to something you can present.
Mistral just raised Europe's largest-ever tech funding round
Mistral AI closed a €3 billion (~$3.5B) Series D led by Samsung Electronics, pushing its valuation above €21 billion (~$24B) — nearly double where it stood a year ago, and the largest equity round ever raised by a European tech company.
The round is stacked: Samsung led, with the EQT-managed Scaleup Europe Fund and existing backer PSG Equity co-leading. New money came from Advent, BlackRock-managed funds, and the government of Luxembourg; existing investors including a16z, ASML, Nvidia, General Catalyst, Lightspeed, and Salesforce Ventures all came back.
Samsung isn't just writing a check — it's deploying Mistral's models across its semiconductor manufacturing business for things like defect detection and equipment optimization, which is arguably a stronger signal than the valuation number itself. CEO Arthur Mensch says Mistral is on pace to cross $1 billion in annual recurring revenue by year-end.
Keep the number in perspective: at roughly $24B, Mistral is still a fraction of OpenAI's ~$852B or Anthropic's ~$965B valuation. This is Europe fielding a genuinely well-capitalized independent player — not yet a peer to the two US frontrunners.
✅ That’s Wrap
SPONSOR US |
Get your business in front of over 30k+ AI professionals
The AI Logbook is one of the fastest-growing AI newsletters, trusted by a community of 30,000+ AI researchers, engineers, and data scientists. |
What We Can Offer: |
Launch an Advertising Campaign
Introduce New Product or Features
Other Business Cooperation
Get in touch today. |
Email our founder Nafis Ahmed at [email protected] |
FEEDBACK
How was your experience with The AI Logbook today?
How was The AI Logbook today?
❤️❤️❤️ Great loved it
❤️❤️ Good, not great
❌ It sucked
Login or Subscribe to participate in polls.
