Grok produced 6,700 sexualized images per hour while other deepfake sites averaged 79. California's AG opened an investigation the same day Musk claimed he saw "literally zero" illegal content. xAI's fix: charge $8 to keep using the feature.
Mira Murati ousted CTO Barret Zoph for "unethical conduct." Hours later, OpenAI welcomed him back along with two other researchers. The competing narratives can't both be true.
Barret Zoph told Mira Murati he was going back to OpenAI. Her response: fire him publicly for "unethical conduct." Hours later, OpenAI welcomed him home. The accusation looks less like discovery and more like revenge.
Fidji Simo said Zoph's return had been "in the works for several weeks." That's a resignation timeline, not a mole timeline. Murati needed a story that wasn't "my CTO quit." She found one.
Meanwhile, California's attorney general opened an investigation into xAI. Grok generated 6,700 sexualized images per hour earlier this month. Two percent appeared to involve minors. Musk said he saw "literally zero."
Stay curious,
Marcus Schuler
Barret Zoph Told Murati He Was Leaving. She Fired Him on the Way Out.
Barret Zoph wanted to go home. Sixteen months after joining Thinking Machines as CTO, he told Mira Murati he was returning to OpenAI. Her response: an all-hands announcement accusing him of "unethical conduct."
The timing tells the story. OpenAI's applications chief Fidji Simo said the return had been "in the works for several weeks." That's not the timeline of a mole. That's the timeline of a resignation. Zoph gave notice. Murati turned it into a termination.
Two other former OpenAI researchers, cofounder Luke Metz and Sam Schoenholz, came back alongside him. The trio had left together. Now they returned together. If Zoph had been leaking secrets, Murati would have known for months. She didn't act until he was already walking out the door.
The "ethics violation" framing reads like damage control. Murati needed a narrative that wasn't "my CTO quit to rejoin the company I left." She got one. Reporter Kylie Robison broke the story.
Why This Matters:
Thinking Machines loses its technical leader less than two years after launch, raising questions about whether Murati can retain top talent
The public accusation suggests internal tension that could spook future recruits and investors watching closely
✅ Reality Check
What's confirmed: Zoph, Metz, and Schoenholz are back at OpenAI. Soumith Chintala (PyTorch creator) replaces Zoph as CTO. OpenAI says the return was planned "for several weeks."
What's implied (not proven): The "unethical conduct" charge was damage control after Zoph gave notice, not evidence of actual wrongdoing.
What could go wrong: Chintala is an infrastructure guy, not a post-training product guy. Thinking Machines just pivoted from consumer AI to tooling, whether Murati intended it or not. The DNA of the company changed in one afternoon.
What to watch next: Whether any other senior staff follow Zoph out, and whether Thinking Machines ships anything beyond API access in the next six months.
California Opens Investigation Into xAI After Grok's Deepfake Flood
Grok generated 6,700 sexualized images per hour during a two-day window this month. Five other deepfake sites combined produced 79. California's attorney general wants to know why.
AG Rob Bonta announced the investigation Wednesday. His office cited evidence of "photorealistic images of children engaged in sexual activity." Hours earlier, Elon Musk had claimed he was "not aware of any naked underage images generated by Grok. Literally zero."
The data tells a different story. Analysis of over 20,000 Grok-generated images found roughly two percent appeared to involve minors. xAI had launched "spicy mode" last summer, explicitly designed for adult content. When the flood became undeniable, xAI restricted image generation to $8/month paid subscribers.
Indonesia blocked Grok entirely on January 10. Malaysia followed the next day. Rep. Zoe Lofgren put it bluntly at a House Science Committee hearing: "Essentially, we're paying Elon Musk to give perverts access to a child pornography machine."
Why This Matters:
State-level enforcement may prove faster than federal action, creating a patchwork of AI content regulations across jurisdictions
The paywall response suggests xAI's strategy is shifting liability rather than preventing harm, which prosecutors will likely challenge
Prompt: A painting of a woman laying on her side in bed, reading with red painted nails and holding up the book to cover half her face, in the style of Maira Kalman.
Google Wants Gemini to Know You Better Than You Know Yourself
Google flipped the switch on personal AI Wednesday. Gemini now connects to your Gmail, Photos, YouTube history, Search history, Maps, Calendar, and shopping data. The pitch: an assistant that actually knows you.
The rollout targets AI Pro and AI Ultra subscribers first. Free tier users will follow. Under the hood, Gemini 3 processes up to one million tokens per prompt, roughly 750,000 words. A technique called "context packing" extracts relevant data subsets rather than ingesting everything at once.
Google VP Josh Woodward offered the requisite caveat: "We've tested this beta version of Personal Intelligence extensively to minimize mistakes, but we haven't eliminated them."
The competitive moat is the point. Microsoft and Anthropic offer limited memory features. Neither has two decades of search history, location data, and video-watching patterns. Google does. The company also confirmed a multiyear deal to power Apple's Siri rebuild, with Apple paying roughly $1 billion annually for a custom 1.2-trillion-parameter Gemini model.
Why This Matters:
Google transforms user data exhaust into a personalization advantage that rivals cannot replicate without similar surveillance infrastructure
The Apple deal locks in distribution to a billion devices while Siri remains Apple's front-end brand, hiding Google's role from most users
Numbers don't speak for themselves. These prompts make them talk.
The So-What Extractor
"Here's a data set: [paste data]. Don't summarize it. Tell me the one finding that would make an executive sit up. Then tell me what question this data can't answer that everyone will ask anyway."
Best on: Claude (handles large data contexts) or ChatGPT with Code Interpreter (can process files)
The Trend Skeptic
"I'm seeing [trend/pattern] in my data. Play devil's advocate: what are three alternative explanations that would produce the same pattern? What data would I need to rule them out? What's the base rate I should compare this against?"
Best on: Claude (rigorous analytical reasoning) or Gemini (can cross-reference with current data)
The Story Finder
"Here's my data: [paste]. I need to present this to [audience]. Find the narrative structure: what's the tension, what's the turning point, what's the implication? Don't give me charts. Give me the story the charts should tell."
Best on: Claude (strongest at narrative structure) or ChatGPT (good at audience adaptation)
Data analysis isn't math. It's translation, from numbers to decisions.
🧰 AI Toolbox
How to Create Talking Avatar Videos with Hedra Character-2
Hedra generates realistic talking head videos from a single photo and audio file. Upload a portrait, add your voice or text-to-speech, and get a video of that person speaking naturally.
The United States conducted a cyberattack against Venezuela that caused a blackout in Caracas during an operation targeting the Maduro government. Senators will question Lt. Gen. Joshua M. Rudd, President Trump's nominee to lead U.S. Cyber Command, about the mission during upcoming confirmation proceedings.
BlackRock Raises $12.5 Billion for AI Infrastructure with Microsoft
BlackRock has raised $12.5 billion through its partnership with Microsoft to fund data centers and energy infrastructure, approaching its $30 billion target. The capital raise underscores massive financial commitments being made to support AI computing demand.
Berlin AI Startup Parloa Hits $3 Billion Valuation
Berlin-based Parloa secured $350 million in new funding at a $3 billion valuation, bringing total capital raised to over $560 million. The startup develops AI-powered customer service agents for clients including Booking.com.
WitnessAI Raises $58 Million for AI Security Platform
WitnessAI, which monitors employees' use of generative AI models, has raised $58 million led by Sound Ventures, bringing total funding to $85 million. The company intercepts GenAI usage and applies protective safeguards.
Trump Jr. Advises Prediction Markets While Father Can Influence Outcomes
Donald Trump Jr. serves as an adviser to prediction platforms Polymarket and Kalshi while Trump Media prepares to launch Truth Predict. Traders can bet on outcomes the president directly influences, raising conflict of interest concerns.
India Supreme Court Orders Tiger Global to Pay Taxes on $1.6 Billion Flipkart Sale
India's Supreme Court ruled that Tiger Global must pay Indian taxes on its $1.6 billion Flipkart stake sale to Walmart in 2018. The landmark decision could reshape how foreign investors structure exits from Indian markets.
Tech Giants Pay Wikipedia for AI Training Data Access
Microsoft, Meta, Amazon, Perplexity, and Mistral have joined the Wikimedia Enterprise program for specialized API access to Wikipedia's content repository. Google was already a member of the program.
Tony Robbins and Other Gurus Launch AI Chatbot Subscriptions
Self-help figures including Tony Robbins and Gabby Bernstein are launching AI chatbots offering personalized advice for up to $99 per month. The trend lets influencers scale personal brands while promising customized guidance.
Anthropic Warns AI Could Widen Global Economic Gap
Anthropic researchers warn that wealthy nations' AI adoption risks deepening inequalities between rich and poor countries. Productivity gains from AI are being unevenly distributed around the world.
ASML Becomes Third European Company to Hit $500 Billion Market Cap
Dutch chipmaker ASML crossed the $500 billion valuation milestone after shares surged following strong TSMC earnings. The semiconductor equipment maker has gained 24% year-to-date, benefiting from AI-driven demand.
AWS Signs Copper Deal for First New US Source in a Decade
Amazon Web Services signed a two-year supply agreement with Rio Tinto to purchase copper from an Arizona mine, the first new domestic copper source in over 10 years. The deal addresses surging demand from AI data center construction.
Bitmine Invests $200 Million in MrBeast's Company
Bitmine, the world's largest corporate ether holder with $13.6 billion in ETH, is investing $200 million in Beast Industries, owned by YouTube star MrBeast. The deal is expected to close January 19.
Spotify Raises US Premium Price to $12.99
Spotify is increasing its Premium subscription by $1 to $12.99 per month for US subscribers, an 8% price hike. This marks the streaming service's first US price adjustment since July 2024.
Wikipedia Turns 25 with 65 Million Articles
The Wikimedia Foundation is celebrating Wikipedia's 25th anniversary, marking growth from 100 pages to 65 million articles attracting 15 billion monthly views. A mini docuseries profiles the volunteer editors behind the encyclopedia.
🚀 AI Profiles: The Companies Defining Tomorrow
Novee hunts vulnerabilities before hackers do. The Tel Aviv startup builds AI that thinks like an attacker, probing networks and applications continuously instead of waiting for annual pen tests. 🔒
Founders
Ido Geffen (CEO), Gon Chalamish (CPO), and Omer Ninburg (CTO) spent decades in Israel's cyber elite, Unit 8200, Talpiot, the Prime Minister's Office. They ran nation-state offensive operations. Now they've bottled that expertise into software. The team hit 32 employees in eight months, split between Tel Aviv and New York.
Product
An AI penetration testing platform that never sleeps. Novee's proprietary 4B-parameter model outperforms Gemini 2.5 Pro and Claude 4 Sonnet by 55% on web exploitation tasks, hitting 90% accuracy where general models plateau at 65%. It maps attack surfaces, chains exploits, finds business logic flaws, and delivers step-by-step fixes. Black-box, gray-box, white-box. Continuous, not episodic.
Competition
Crowded space getting louder. Pentera automates validation. Cobalt and HackerOne run crowdsourced testing. Tenzai raised $75M chasing similar ground. Geffen frames Novee's ambition like Wiz for pen testing: comprehensive, continuous, impossible to ignore. The edge is accuracy. Finding real vulnerabilities beats generating false positives.
Financing 💰
$51.5M in under eight months: $8.5M seed in May 2025, $33M Series A in September, $10M venture debt in December. YL Ventures, Canaan Partners, and Zeev Ventures led. One of the fastest funding trajectories in offensive security. Customers include K Health, HiBob, JetBrains, and Telit.
Future ⭐⭐⭐⭐
AI-powered attacks accelerate. Human pen testers can't keep pace. Novee's bet: enterprises need machine-speed defense against machine-speed offense. The risk is trust. Security teams don't hand network access to tools they can't verify. But if AI pentesters become standard, Novee wrote the playbook first. 🛡️
Tech translator with German roots who fled to Silicon Valley chaos. Decodes startup noise from San Francisco. Launched implicator.ai to slice through AI's daily madness—crisp, clear, with Teutonic precision and sarcasm.
E-Mail: marcus@implicator.ai
The Trump administration approved exports of Nvidia's H200 processors to China on Tuesday. Hours later, Beijing told its tech companies to stop buying.
