Welcome back! Cybersecurity researchers have uncovered an industrial-scale catfishing network operating across 28 dating apps. The network, powered by an enterprise Anthropic API connection, generates over 100,000 requests per day and deploys autonomous AI personas to hold multi-week conversations, stage fake video calls with ambient noise, and extract payments.
The dream was an AI assistant managing your inbox. The reality is an AI bot running a "swipe right, wire money, ghost" scam on your heart.
What If The Best AI Insights Come From Going Technical?
For anyone who thinks AI content needs to stay fast, light, and high-level: why is the raw, complex stuff where the real breakthroughs happen?
I took a gamble and started a new series on my YouTube channel. The goal was to bring on other creators who are doing genuinely interesting things with AI and have them break down their actual workflows step-by-step.
In the first episode, Bilawal Sidhu walked me through the open-source "God's Eye" system he built. Going into it, I was nervous. It gets pretty nerdy, dives deep into the technical weeds, and strays completely from the usual format I post. Sometimes when you take a risk like that, it flops.
And sometimes, it REALLY works out. There is so much to learn from seeing how someone puts these tools together in real life, and I'm pumped to do a lot more of these conversations. Big thanks to Bilawal for nerding out with me!
If you want to check out the first episode, you can watch it here:
My POV: It’s one thing to know the tools that exist. But understanding how actual builders connect them under the hood to solve real problems and watching someone build in the trenches—that’s where real skill transfer happens.
Hit reply and let me know: do you prefer quick feature overviews, or deep-dive workflow breakdowns?
— Matt
OpenAI Eyeing $1.2 Trillion Valuation in Pre-IPO Talks
OpenAI has held early discussions with major institutional investors for a massive funding round that could value the company at roughly $1.2 trillion.
The trillion-dollar benchmark: Venture funds expressed interest, rather than OpenAI initiating the talks. If finalized near this target, the round would dwarf OpenAI’s previous funding round, where it raised $122 billion from Amazon, Nvidia, and Softbank at an $852 billion valuation.
A growth story: OpenAI has reached more than 1 billion active users, while upwards of 200 million businesses use its models. Its latest model, GPT-6 Astra, surpassed expectations and is one of the world’s most intelligent models.
Capital surges past safety warnings: Institutions like Apollo Global Management are considering expanding credit facilities to partners like SoftBank to fuel deeper equity stakes in OpenAI. The capital injections are reaching historic peaks at the same time safety researchers and rival labs issue increasingly stern warnings about model risk.
The bigger picture: OpenAI keeps growing, and investors want in. Calls for a pause—often coming from within the labs themselves—have yet to slow investors and venture funds’ infusion of capital.
The Great AI Divide: Free-Market Speed vs. Precautionary Brakes
A growing rift has emerged between hardware suppliers pushing for maximum deployment velocity and frontier AI labs advocating for structural safety controls.
Jensen Huang says trust the market: Speaking at Salesforce’s Dreamforce conference, Nvidia CEO Jensen Huang pushed back against calls for AI legislation, declaring that "safety is an engineering problem, not a legal one." Huang argued that free-market forces naturally penalize companies that release faulty products.
DeepMind and Anthropic urge caution: On the other end of the spectrum, Google DeepMind co-founder Shane Legg launched the DeepMind Institute to study AGI risks, warning that advancing model capabilities must not outrun safety controls. Legg aligned with Anthropic CEO Dario Amodei’s call to consider slowing down frontier model releases.
Model transparency in decline: DeepMind researchers specifically pointed to OpenAI’s recent GPT-6 Astra release as evidence of a concerning downward trend in model transparency. The debate highlights a fundamental disagreement over whether risks can be patched after deployment or must be solved prior to release.
The bigger picture: This tension will dictate how the next generation of super-intelligent systems reaches the public. If hardware providers and research labs remain at odds, regulatory bodies may be forced to intervene and set mandatory safety benchmarks for all developers.
Your New AI Teammates Get Work Done. No Prompting Needed.
Imagine opening your laptop Monday morning and finding that your SEO report, competitor updates, meeting prep, and next campaign recommendations are already there, waiting for you. With Optimizely’s new Virtual Teammates, you don’t have to imagine.
Join Optimizely on September 29 to see how you can:
Add extra capacity to your team by hiring Virtual Teammates with role-based responsibilities
Use them to keep work moving and collaborate on execution, always completing work to your standards
Get a full audit trail of everything they do, and configure when human reviews and approvals are required
Want to meet your Virtual Teammates? Save your webinar spot today.
Turn Workspace Context Into Daily Briefings
Via Dreambeans
Dreambeans is an experimental morning briefing application built by Google Labs. Working overnight, it synthesizes data across connected Google Workspace services into an illustrated, visual story deck highlighting key updates for your day ahead.
How you can use it
Receive AI-illustrated visual summaries of key project milestones
Highlight urgent tasks and schedule conflicts before starting your day
Connect Gmail, Calendar, Photos, and Search context seamlessly
Pricing: Free
Track Your Brand in AI Answers
Via Yolando
Yolando is a marketing analytics platform designed to track how platforms like ChatGPT and Perplexity cite and rank your brand across target queries.
How you can use it
Monitor brand citation frequency across LLM search engines
Identify prompts where competitors are recommended over your product
Generate prompt-optimized content strategy recommendations
Track accuracy and sentiment in AI-generated answers
Pricing: Paid
Jobs, announcements, and big ideas
Security researchers using Anthropic's Claude hacked into OpenAI.
OpenAI launches Astra for Law to bring GPT-6 intelligence to legal research and workflows.
xAI adds full-duplex Voice Mode to Grok for real-time natural conversations.
Google Labs expands its CC AI agent to help families and households coordinate schedules.
Anthropic introduces the Life Sciences Verification Program to expand AI access for biology research.
Runway introduces Enhance Frame Rate to convert videos up to high broadcast rates.
Mastercard teams up with startup Alchemy to let AI agents make payments on users’ behalf.
Midjourney updates its Alpha web interface to introduce official Korean language support.
It’s been a hectic week of news. I give you the tl;dr on pacing AI, plus much more ⤵️
That’s a wrap! See you next week for more.
