The Kimi conversation refuses to stay in one lane. Friday it was about enterprise adoption. Yesterday it was about Google Cloud being the unexpected beneficiary. Today it's about cybersecurity capability and what happens when the U.S. response to Chinese AI competition is to lock things down tighter. The irony is that the people arguing for more restrictions may be doing more damage than the models they're worried about.
01
Kimi K3 just passed a cybersecurity stress test that "Fable" completely failed
Vercel CEO Guillermo Rauch posted internal evaluation results comparing three models on cybersecurity tasks. Kimi K3 ranked as top-tier, calling out the benchmark-overfitting chatter directly: these were stealth evals, not public benchmarks, and the model held up. A model called Sol outperformed everyone on raw capability but at significantly higher cost. A model called Fable refused to complete the run at all, apparently too locked down to engage with even defensive security hardening tasks. ---
Why it matters: A model that won't help your security team do their job isn't safe, it's useless. If your company is evaluating AI for defensive cyber use and you're defaulting to the most restrictive option, you may be handing an advantage to adversaries who aren't applying the same brakes. The Sol/Kimi split on cost versus capability is exactly the tradeoff every security team will face in the next procurement cycle.
Box CEO Aaron Levie: locking down AI models is how America loses
Box CEO Aaron Levie posted a direct argument against tightening U.S. AI export controls or restricting model access. His read: China has already crossed the threshold where it can compete at near-frontier levels, and that was true even before Kimi K3. Trying to slow them down by restricting your own ecosystem just guarantees the U.S. loses the broader race. ---
Why it matters: This is the core policy debate happening right now, and the outcome will directly shape which models your company is allowed to use, which vendors can serve you, and how fast the rest of the world catches up. If Levie's right, the companies lobbying for tighter controls aren't protecting American AI. They're protecting their own market position.
OpenAI's "ChatGPT Work" is now trying to replace your whole productivity stack
Thibault Sottiaux posted a breakdown of what ChatGPT Work covers: building and hosting sites, managing email, summarizing documents, and creating docs, sheets, and slides. It's included in Plus, Pro, Business, and Enterprise plans and available on mobile. ---
Why it matters: OpenAI is no longer selling a chatbot. It's selling a competing answer to Microsoft 365 and Google Workspace. If your company pays for both, someone in finance is going to ask why. The more interesting question is whether the people who use these products daily will actually switch, or just get ChatGPT added to a stack that keeps growing.
Zara Zhang on the real reason enterprise AI stalls
Zara Zhang posted a single-sentence observation that's been circulating: the biggest barrier to enterprise AI adoption is that the people who understand AI don't understand the business, and the people who understand the business don't understand AI. ---
Why it matters: This is why your company's AI transformation is probably being run by either someone who can demo a model but can't read a P&L, or a business leader who approved a six-figure contract without knowing what an API is. Neither ends well. The companies that close this gap first, by hiring for both or actually cross-training, are going to move faster than everyone else.
Swyx, who runs Latent Space, posted that critics dismissing Europe's AI scene are missing something real: some of the world's top AI engineers are European, and the talent is there if you look for it. He noted his conference is running what he called the most competitive global arena for AI talent right now. The post is thin on specifics, and the engagement reflects that. Worth a flag if you follow the European AI talent question, but not a lot to build on here.