🛎️ Crossed The Line

Good Morning, AI Enthusiasts!

AI stopped looking like a tool. It discovered, created, built, and even faced its first “who’s responsible?” moment.

FUTURE

The Creativity Line Crossed

👀 What's happening: Wharton statistician Edgar Dobriban says GPT-5.6 spent 90 minutes cracking a problem that sat open for 20 years. It involves Benjamini-Hochberg, a method used across science to keep false positives out of big data, from gene screens to brain imaging. It was only ever proven to work when data is independent; everyone assumed it held when data is correlated too, as real data usually is. GPT-5.6 found the case where it doesn't. A month earlier the same job ran GPT-5.5 for 20 hours and failed. Dobriban and Berkeley's Will Fithian checked the work and posted the code. The method is cited over 130,000 times, and nobody had ever found the crack.

🌍 How this hits reality: Yesterday's proofs, the graph theory conjecture, the string theory notes that we talked about could still be dismissed as superhuman stamina, brute-forcing paths humans already knew. This can't. The model built a structure no one had seen, stitching asymptotics to a numerical certificate in a way statisticians called atypical. Going from 20 failed hours to 90 winning minutes in one model generation isn't speed, it's a phase change across the line humans called creativity.

🛎️ Key takeaway: Yesterday's breakthroughs could still be called brute force. This one was invention, and it took ninety minutes.

ETHICS

Grok Plays The Victim

👀 What's happening: SpaceXAI is suing Terry Harwood, a South Carolina man, claiming he bypassed Grok's safeguards to turn nonconsensual photos into CSAM and distribute them. Harwood was arrested in February on eight CSAM felony counts and belongs in prison. But read what xAI is actually claiming: not harm to the children in those images, but harm to itself. The suit says Harwood caused the company "legal risk and reputational damage," and demands he cover xAI's losses and its future legal bills when victims sue. It's the first time xAI has sued over a Grok deepfake.

🌍 How this hits reality: Here is who built this. Grok shipped the spicy mode. Grok added the image editor. A flood of sexualized deepfakes followed, some of minors, and teens sued Grok in March. Every safeguard Harwood supposedly "circumvented" was bolted on after the tool was already generating this. So xAI made the machine, profited from the mode, and now sues a user to move the blame, and the bill, onto the person who pulled the trigger it built.

🛎️ Key takeaway: xAI built the gun, shipped it without a safety, and is now suing the shooter for making the gun look bad.

ROBOTICS

Neo Launched the Most Human Hand Yet

👀 What's happening: Robotics firm 1X gave its home robot Neo the most human-like hand in robotics, and it earns the label from the inside out. Instead of packing motors into the fingers like everyone else, 1X put them in the forearm and runs cables down to the joints, exactly how tendons pull a human hand. Each hand gets 25 degrees of freedom against a human's 27, but the point is the wiring, not the count. The fingers grip odd shapes, sense a slipping object, and hyperextend past what human digits can. IP68 rated, Neo can wash its own hands and plug itself in.

🌍 How this hits reality: Dexterous robot hands aren't new. Shadow matched these joint counts years ago, Tesla's Optimus runs 22, Sanctuary around 20. But every one of them was built backward from the result, motors jammed into the knuckles to reproduce what a hand does. 1X started from how a body actually makes one, muscle in the forearm, tendon doing the pull. Copy what a hand does and you get a claw that imitates it. Copy how a hand is built and you get something that is one.

🛎️ Key takeaway: Everyone else built a hand that acts human. 1X built one that's made like a human, and those turn out to be very different things.

NEW LAUNCH

Murati's Anti-Frontier Bet

👀 What's happening: Thinking Machines, the startup led by former OpenAI CTO Mira Murati, released its first model, Inkling. It's open-weight, reasons across text, images, audio, and video, outputs code and structured data. The company says outright it is "not the strongest overall model available today, open or closed." The pitch is customization instead: download it, then fine-tune it through Tinker, the company's paid platform. In one joint project, hedge fund Bridgewater trained an open model on its own expertise and reportedly hit 84.7% on financial reasoning at a fourteenth of the cost, beating closed rivals. Thinking Machines built all this in nine months.

🌍 How this hits reality: Admitting your model isn't the strongest isn't humility here, it's the strategy. If you can't win the frontier, you argue the frontier stopped mattering. But the customization gospel has a tell: Inkling ships as a starting point, not a product. Safety becomes the customer's problem, and the fine-tuning it needs takes an ML team most buyers don't have. The model everyone can download quietly serves the few who can afford to finish it.

🛎️ Key takeaway: When you can't build the smartest model, you sell everyone on the idea that smartest was never the point.

TOGETHER WITH MYCLAW

Run Best-in-Class AI Agents — In Seconds, Always On

MyClaw is a best-in-class AI agent platform that brings together OpenClaw, Hermes Agent, and commercial infrastructure to help businesses launch intelligent agents that understand goals, use tools, automate workflows, and deliver finished results.

We already host OpenClaw and Hermes — and we're just getting started. Claude Code, Codex, and every agent worth running are coming next. One home for all of them, always on.

Stop babysitting infrastructure. Spin up your agent, walk away, and let it work around the clock.

MyClaw — Your Agents, Hosted Right.

DAILY TL;DR

• Microsoft is training sales teams to pitch its own AI stack against OpenAI, Anthropic, and Google, making the model war inside enterprise sales explicit.
• OpenAI launched a $230 Codex Micro keypad for managing coding agents, giving agentic work its first weird little hardware command center.
• Suno was hit by leaked code suggesting it scraped millions of songs from YouTube, Deezer, Genius, and other music sources for training.
• OpenAI introduced GPT-Red, an internal red-team model used to attack its own systems and harden GPT-5.6 against prompt injections.
• OpenAI employees donated to a rival AI-safety super PAC opposing a pro-industry group backed by company president Greg Brockman.
• Apple won approval to launch Apple Intelligence in China through Alibaba’s Qwen, giving its delayed AI rollout a path into a key market.
• Nvidia introduced new Jetson Thor computers for robotics and edge AI, pushing Blackwell-powered physical AI into smaller, cheaper deployed systems.

READ MORE

Let the Future Come to Your Inbox

Stay ahead without drowning in information. We turn the most important signals across AI, tech, marketing, and future products into 5-minute reads you can actually finish.

• AI Secret uncovers what really matters in AI
• Bay Area Letters decodes tech and business shifts from Silicon Valley
• Robotics Herald tracks how robots move from labs into daily life
• Marketing Secret breaks down real growth and go-to-market playbooks
• The Hardwire explores hardware, consumer tech, and what’s coming next

TOGETHER WITH US

AI Secret Media Group is the world’s #1 AI & Tech Newsletter Group, reaching over 2 million leaders across the global innovation ecosystem, from OpenAI, Anthropic, Google, and Microsoft to top AI labs, VCs, and fast-growing startups.

We've helped promote over 500 Tech Brands. Will yours be the next?

Email our co-founder Mark directly at mark@aisecret.us if the button fails.