🛎️ KIMI‘s New K3 Comes for Claude

Good Morning, AI Enthusiasts!

The future is open-source, regulator-approved, spreadsheet-optimized, and possibly backdoored.

NEW LAUNCH

KIMI‘s New K3 Comes for Claude

👀 What's happening: Moonshot AI, the Chinese lab behind the Kimi assistant, unveiled K3, a 2.8-trillion-parameter model with a million-token context window, and said it will fully open-source it on July 27. It ranks third on Artificial Analysis's intelligence index, behind only Fable 5 and GPT-5.6 Sol, and API pricing lands around Anthropic's Sonnet tier. It arrives months after DeepSeek crossed the trillion mark, making trillion-scale models look almost standard in China now.

🌍 How this hits reality: A trillion parameters is becoming the price of admission, and the fee is brutal. K3 is a well-rounded model, near the top on most benchmarks yet best-in-class on almost none. But the real problem is not building it but running it: the compute to serve a model this size is so scarce he expects access to sell out within days. Open-sourcing the weights costs Moonshot nothing it cannot spare. The electricity and chips to actually run them stay locked up. The barrier just moved from talent to power plants.

🛎️ Key takeaway: The weights are free to download and impossible to afford to run, which is the real shape of an AI race only nations and giants can enter.

SEARCH

The EU Is Fighting Yesterday's Google

👀 What's happening: The EU is ordering Google to hand rivals two keys to its kingdom. Starting in January 2027, Google must share anonymized search optimization data with OpenAI and other AI search services under a regulated pricing formula. By July 2027, it must open 11 Android functions, allowing rival assistants to answer voice commands, find places, and book rides.

🌍 How this hits reality: Brussels is applying the monopoly map of the search era to an AI market nobody has won. Google owns immense distribution and data advantages, but Gemini has not established anything close to Search’s dominance, with OpenAI may already be closer to the next gateway. The Digital Markets Act was built to open mature gatekeepers. Here it risks tilting a new market before dominance is even settled. Forcing Google to strengthen OpenAI and other challengers converts yesterday’s moat into a subsidy for whoever captures tomorrow’s gateway.

🛎️ Key takeaway: Brussels is building a key for a lock AI is quietly moving, and handed it to the company already holding the new one.

VIDEO

The AI Crept Into 300 Netflix Titles

👀 What's happening: Netflix told investors roughly 300 of its titles have used generative AI, most in post-production, calling the tools a way to deliver "higher quality output more quickly and at a lower cost." Co-CEO Ted Sarandos said The American Experiment docuseries carries 17 minutes of "AI-enhanced footage," produced twice as fast and at half the cost, and argued the shots would otherwise have been cut as unaffordable. Other cited titles used AI for crowds and historical battle sequences.

🌍 How this hits reality: Sarandos framed it as artistic rescue that AI lets productions afford shots they would have dropped. That is the cover story. The number is the story. Three hundred titles is not experimentation, it is a default pipeline, and a CEO who needs a full earnings call to justify it is answering to cost, not craft. "Enhanced crowds" and "worldbuilding shots" are where the savings hide today. Next year they may move up the call sheet.

🛎️ Key takeaway: When the boss spends a quarterly call explaining why the AI is fine, the decision was made on a spreadsheet, and the reasons came after.

SECURITY

Open Weights Are Not "Open"

👀 What's happening: Katie Paxton-Fear, a cybersecurity lecturer at Manchester Metropolitan University, says she used ten training examples, one hour, and less than $100 to backdoor an open-weight model. It then produced remotely exploitable code across unfamiliar prompts and domains. A separate poisoned model quietly emailed out drug discovery data. Nothing crashed. No malicious instruction arrived at runtime. The attack was already inside the weights.

🌍 How this hits reality: A model is billions of raw numbers, too many for any human to read, and nothing about the file reveals what it will do. Traditional software can be pulled apart and traced; we have mature tools to spot a poisoned dependency. For models, that layer barely exists. A corrupted reuploaded one does not crash or throw an error, it just quietly steers a decision, and the industry has almost no way to catch it.

🛎️ Key takeaway: We are wiring billions of unreadable numbers into real decisions faster than we can build any way to tell when they have turned on us.

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

• SoftBank fell 9% as Asian chip stocks followed Wall Street’s AI sell-off, showing investors are questioning AI infrastructure returns.
• OpenAI admitted GPT-5.6 can sometimes delete files without permission, calling the rare data purges an “honest mistake.”
• DoorDash can now be ordered from the command line, another small sign that agent-style commerce is moving into everyday services.
• SpaceXAI is facing Memphis backlash as Musk’s AI data-center buildout strains local power and pollution politics.
• New York governor Kathy Hochul said the state is using AI to analyze every rule on its books, bringing AI into regulatory cleanup work.
• MLB banned teams from using dugout iPads for AI-powered strategy calls, cutting off AI help for substitutions, pitching, and defensive decisions.
• Roblox is adding mobile AI game creation, letting users turn text prompts into playable Roblox experiences directly from a phone.
• 1Password launched a Claude integration that lets the AI use saved logins without exposing passwords or MFA codes to the model.

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.