In partnership with
Welcome Automaters, 👋
OpenAI is claiming it just cracked one of the absolute hardest math puzzles on the planet, but before they could even pop the champagne, the whole thing spiraled into a messy, high-stakes drama on social media. We’re talking about a rival lab, a furious math professor, and wild allegations of backroom Ultimatums.
What Actually Happened?
OpenAI officially announced it solved a core piece of the Navier-Stokes problem. And in case you don't know what that is, here’s what it’s about:
Navier-Stokes problem is one of the seven iconic Millennium Prize Problems. They’re so notoriously brutal that the Clay Mathematics Institute put a $1 million bounty on each back in 2000.
The Rarity: Only one other problem on that list has ever been solved in the last quarter-century.
Wait, What Even Is Navier-Stokes? Think of it as the ultimate physics cheat code. It's the mathematical framework governing how fluid and air move through our world.
The core question asks something seemingly simple: Are there bizarre edge cases where these rules break down and predict something impossible in real life? For decades, nobody could prove it either way, until OpenAI says its AI just did.
Then enter NYU mathematician Tristan Buckmaster and Levent Alpöge (an Anthropic researcher working on this independently). According to their statements, they were already closing in on a solution using a mix of AI tools, including Claude and OpenAI's Codex and Astra models.
Then things got spicy:
Buckmaster publicly alleged that OpenAI suddenly rushed its own announcement right after discovering his team was getting close.
He directly asked OpenAI if its system was trained on his private Codex drafts. He claims he got total silence in return.
Buckmaster says OpenAI hit him with two absurd options: one, publish a watered-down, partial version of his work before OpenAI dropped the full solution the next day. Or write a solo paper crediting OpenAI's tool, while wiping his collaborator Alpöge's name off the paper entirely, because of his Anthropic tie. He of course turned down both.
On the other hand, OpenAI’s Sebastien Bubeck and CEO Sam Altman immediately fired back on X. Bubeck flat-out denied ever asking to remove Alpöge’s name.
So right now, it’s a complete he-said-they-said situation playing out in real time with zero hard receipts released by either side. The only detail everyone actually agrees on? OpenAI officially announced it won't claim the $1 million cash prize.
The Takeaway: Conquering a century-old mathematical mystery should be an easy, unanimous win for humanity. Instead, it devolved into a messy, public feud on X, which tells you exactly how competitive the AI research race has gotten.
And get this: AI is officially pushing the absolute boundaries of deep science, but considering training these models requires scraping other people's work and expertise, deciding who actually gets the gold star is turning out to be way messier than the math itself.
Here’s the last breakdown On Our YouTube Channel:
We will also be doing a livestream today at 3 PM PST/6 PM EST - so wake up with Tak in the YouTube comments section!
Here's what we have for you today
🤯 Claude Users Are Waking Up to Mysterious Token Drains, and Anthropic Isn't Fully Explaining Why
Imagine dropping $200 a month for top-tier AI power, only to sit back and watch your usage meter leap from 45% to 55% in real time while you’re literally doing zero work.
Meet Grant De Swardt. He's an AI consultant out in East Sussex, U.K., and on August 4th, he noticed his Claude Max 20x account was burning through tokens like a runaway freight train. He was not working. He paused every active task. Yet, the usage meter just kept climbing. Creepy, right?
When De Swardt reached out to Anthropic, they could not (or simply would not) provide an itemized receipt showing what actually devoured his balance. Behind the scenes, however, the alarm bells were clearly ringing. Anthropic quietly suspended his account, wiped his active sessions, and handed over a modest £44.49 refund.
The official diagnosis? Investigators revealed that a compromised session key was hijacked to mint sneaky, unauthorized Claude Code OAuth tokens. Basically, a hacker manufactured a digital skeleton key to his account, turned it into a covert token-siphoning operation, and used his paid access to service total strangers.
Because Anthropic tracks overall token consumption rather than itemized usage, this kind of freeloading could easily fly under the radar for months.
De Swardt took his story to Reddit, and the comment section instantly blew up with users sharing eerily identical nightmares:
One user reported their account randomly upgraded itself without permission, charged their credit card, and maxed out usage from 0% to 100% untouched.
Another watched their balance jump from 0% to 49% in just 12 minutes after running two basic prompts and a simple web search.
A third subscriber had their entire daily quota completely wiped three days in a row without sending a single prompt, eventually prompting them to file a formal GitHub issue.
So Who’s the Real Villain Here?
It turns out bad actors are actively deploying infostealer malware. This nasty software sneaks onto your personal computer through malicious ads, sketchy downloads, or compromised file links. Once inside, it quietly scrapes your saved passwords, login credentials, and active browser sessions straight off your machine. No massive server breach at Anthropic required!
Anthropic actually sent warning emails to a handful of impacted users, confirming that bad actors were using infostealers to hijack session data and drain accounts. In those specific cases, the company logged out the compromised sessions, invalidated tokens, issued refunds, and advised users to clean their devices.
Here’s the wild kicker: De Swardt never received that warning email.
He maintains there is no evidence his computer was ever infected, leaving him with zero clear answers on how the intruder got inside. When asked how regular users can spot or audit suspicious account activity, Anthropic declined to comment. Messy right?
Well right now, platforms lack granular tracking tools for users to monitor real-time token allocation. And until transparent, itemized usage logs become standard, here’s how to stay safe right now:
Clear out idle logins: Log out of your active Claude sessions when you are not actively using them.
Audit your extensions: Remove any browser add-ons or downloaded software you do not 100% trust.
Run a thorough system scan: Check your local machine regularly for hidden malware strains.
PS: Don’t forget to catch us on YouTube for a visual breakdown later today!
The best influencer marketing advice never comes from a report
Think about the last thing that got you fired up about your job this year. Odds are it didn't come from a report. It came from someone who does the job, saying what happened when they tried it.
So we've filled a day with people like that.
Maya Shaff, ŌURA. Leah Walker, Adobe. Georgia Humphries, Stanley 1913. Enara Roy, Halfday. Tyler Vaught, Edelman. Josh Rangel, Ogilvy. Sarah Whittle, ex-Crocs and Duolingo.
And that's just the first wave of speakers.
Return on Influence Festival '26 is a free, virtual day dedicated to influencer marketing. Like any good festival, there are stages. But no mud, no queues, and absolutely no porta-potties.
Sign up today, and the rest of the lineup lands in your inbox first.
🧱 Around The AI Block
🤔 Meta debuts its Muse AI agent. Will consumers trust it?
🤯 Chrome is now shipping updates every 2 weeks as AI changes the security landscape.
🦾 AI use now factors into promotions at tech firms.
👩⚖️ Anthropic sued over Claude Max pricing claims.
🤑 French AI company Mistral hits $24 billion valuation in funding round.
👏 AI will help find cure for cancer ‘within our lifetimes’, says Arm Holdings chief.
😎 Andrew Garfield takes on Sam Altman in creepy first teaser for Artificial.
🤝 Google Cloud races to catch up in the AI deployment wars with Accenture deal.
🛠️ Trending Tools
For the Music Producers & Audio Creators: Lyria 3.5 is Google DeepMind’s flagship AI music model, generating full-length, high-fidelity stereo tracks from simple text or image prompts. It handles complex musical structures; including multi-verse layouts, choruses, and bridges, while synthesizing expressive vocals and instrumentation up to 3 minutes long. It offers musicians, video editors, and content creators an effortless way to prototype studio-quality songs and exports.
For the Information Researchers & Fact-Checkers: Wikis.ai prompts multiple leading AI models simultaneously and displays their raw, independent answers in a side-by-side comparison grid. It highlights shared consensus points alongside areas of divergence, making it easy to spot potential hallucinations or distinct model perspectives without juggling multiple browser tabs. It provides researchers and writers with a transparent workspace to evaluate complex topics across distinct AI architectures.
For the Social Media Managers & Archival Leads: Telegram Video Downloader extracts raw video, image, and audio files directly from public and private Telegram channels. It preserves original media quality during export without requiring user registration or account creation. It gives media archivists and social producers a straightforward way to download community content for offline viewing or editing.
🤖 AI Workout Of The Day: Crafting High-Impact, Emotionally Resonant Songwriting Frameworks
Great lyricism isn't just about rhyming words over a beat, it’s about building a compelling narrative arc and invoking vivid emotional resonance.
When song lyrics rely on generic tropes or awkward phrasing, they fail to connect with listeners or fit the rhythmic cadence of the music. A well-crafted songwriting prompt guides the flow, rhyme scheme, dynamic shifts, and thematic imagery of a track. By establishing clear emotional anchors, structural boundaries, and stylistic influences, you transform simple musical concepts into memorable, record-ready anthems.
💡 Prompt To Try:
Act as a Grammy-winning songwriter, lyricist, and top-line melody arranger across genres. Your goal is to write complete, execution-ready song lyrics based on the following creative parameters:
* Song Title or Working Concept: [INSERT TITLE OR CONCEPT]
* Genre & Artist Vibe: [e.g., Indie Pop / Phoebe Bridgers, Cinematic R&B / Frank Ocean, Modern Country / Luke Combs]
* Core Theme & Mood: [e.g., Nostalgic heartbreak, High-energy empowerment, Melancholic reflection]
* Primary Musical Structure: [e.g., Verse 1 - Pre-Chorus - Chorus - Verse 2 - Pre-Chorus - Chorus - Bridge - Chorus - Outro]
* Key Imagery, Phrases, or Story Anchors: [INSERT SPECIFIC PHRASES, MEMORIES, OR METAPHORS TO INCLUDE]
* Target Tempo & Energy Curve: [e.g., Mid-tempo 90 BPM building from intimate acoustic to an explosive final chorus]
Please deliver a complete Lyric Execution Suite structured into the following 4 distinct sections:
1. THEMATIC & VISUAL CONCEPT BOARD: Detail the core emotional narrative and 3 sensory motifs (e.g., visual imagery, sound metaphors, tactile feelings) used throughout the song.
2. Provide the full lyrics formatted with clear structural tags [Verse 1, Pre-Chorus, Chorus, Bridge, Outro].
* Include dynamic performance cues in brackets [e.g., (Vocals drop to a whisper), (Full band kicks in), (Belting high note)] to guide the vocal delivery and emotional build.
* Maintain a consistent rhyme scheme (e.g., AABB, ABAB) and rhythmic meter tailored to the genre.
3. RHYTHMIC & MELODIC GUIDE: Highlight the core hook of the Chorus and provide guidance on phrasing, syllable counts, or cadences to make the main melody stick in the listener's head.
4. SECONDARY VARIATION (Optional Hook Alternative): Provide 1 alternative version of the main Chorus hook with a slightly different emotional angle or lyric tweak for producer options.
TONE & EXECUTION GUIDELINES:
* Focus on "showing, not telling" through vivid imagery and authentic human emotion rather than cliché rhymes.
* Ensure rhythmic scannability so the lyrics naturally fit standard song structures.
PS: If there’s a story or special kind of inspiration behind the song, remember to add that to the prompt, and be as detailed as you can be. The more context the better the result will be.
Is this your AI Workout of the Week (WoW)? Cast your vote!
That's all we've got for you today.
Did you like today's content? We'd love to hear from you! Please share your thoughts on our content below👇
What'd you think of today's email?
Your feedback means a lot to us and helps improve the quality of our newsletter.
