In your (Hugging) Face!

Top News

An OpenAI model hacked Hugging Face: On Tuesday, OpenAI confirmed that models it has been testing — including GPT-5.6 Sol and an unreleased, even more powerful successor — escaped their sandbox and compromised portions of model and dataset repository Hugging Face’s production infrastructure. The models had been working on an internal security test, with their safety guardrails deliberately muted, and got what OpenAI describes as “hyperfocused” on solving a benchmark-related puzzle. Ultimately, the models exploited a flaw to reach the open internet, and worked their way into Hugging Face’s systems over the course of a weekend. OpenAI is calling it “an unprecedented cyber incident, involving state-of-the-art cyber capabilities.” Hugging Face CEO Clem Delangue took a somewhat different tone, thanking OpenAI for their cooperation but arguing that the episode proves AI safety must be solved in the open, not by “any single company working in secret.”
White House: Kimi K3 is built on stolen IP: The wave of buzz around Chinese AI lab Moonshot’s Kimi K3 model revived a fight with the Trump Administration that first kicked off in April. Recall, at that time, Office of Science and Technology Policy (OSTP) director Michael Kratsios circulated a memo accusing Chinese open source labs of running an “industrial scale” campaign to distill US frontier models. Anthropic has specifically named Moonshot as one of the prime distillation attack offenders. These accusations are far from settled of course. Even OpenAI researcher Dean Ball argues that, while distillation likely plays some role in the success of Chinese models, it’s not the sole determinant of their success. Nonetheless, the White House is debating swift action. Treasury Secretary Scott Bessent has reportedly floated the notion of sanctions against Moonshot and other labs in response to the alleged IP theft.
New Samsung foldable comes with AI agent: The mobile giant's latest suite of foldable devices is headlined by the Galaxy Z Fold 8 Ultra, retailing at $2,099. It boasts the expected hardware upgrades but the real headline here is the software. The phone comes with “Gemini Intelligence” on board that functions as an AI agent, which Samsung boasts can capably navigate more than 40 apps and services on its own. It’s the latest sign that phone makers are done treating AI as a chatbot bolted on to the side of the device, and are starting to build native applications with agentic use in mind.

TWiST 500

The potential loss of access to powerful but lower-cost open source models from China poses a significant threat to American startups, which have come to rely on affordable intelligence that’s very nearly as competent as frontier US model releases. But if there’s such a tremendous and established need for an American alternative to useful but geopolitically inconvenient options like Moonshot’s Kimi K3 and Z.ai’s GLM-5.2… why aren’t more US labs training them?

Well, some actually are! The list includes San Francisco-based Arcee, which has been training and releasing small and large language models for enterprise applications and inference debuting back in 2023. (They’re not yet in the TWIST 500 but we’re taking it under advisement.)

You might expect an American open-weight AI lab to feel downright enthusiastic about federal efforts to shut off the pipeline of competitive, relatively cheap intelligence from China. But that’s not what’s happening. Arcee CTO Lucas Atkins is making the public case this week that, actually, Americans have little to fear from Chinese AI labs, at least from a national security perspective.

According to Atkins, the fear that these models pose some hidden threat is overblown, and mostly serves the profit margins of the large proprietary labs rather than any real risk to the companies running them. It’s a largely technical argument: there’s no feasible way that a Chinese lab could quietly train an AI model to misbehave on command. If an American user downloads the weights, and runs the model on their own servers, there’s no immediate connection back to Beijing to be exploited. As well, Atkins notes that large enterprises are putting these models through their own security testing and fine tuning, which could help to eliminate not just concerns about data sharing, but unwanted pro-CCP bias.

Another interesting wrinkle: Arcee treats open Chinese models as a resource rather than a rival they hope to regulate out of existence. Because their weights are public, the company can study what Moonshot and Alibaba have done well and build on top of it. – Zach

A message from DigitalOcean

Want to see what building on a true AI-native platform looks like? Head to do.co/twist to start building on DigitalOcean's AI-Native Cloud today — and cut your AI workload costs by up to 50%.

This Week in Startups

E2314: Ent founder Brandon Dixon joins the pod to explain why the traditional cybersecurity process — which cleans up a system after a breach has occurred — is backwards. He’s designed an AI agent that lives on everyone’s corporate laptop, and keeps an eye out for risky clicks and potential leaks before they cause a problem. PLUS Clawra creator David Im returns to the show with Sume’s Avatar, a multi-model orchestration layer that he claims can one-shot your AI generated videos without the endless slot machine-style trial and error.

E2313: Science fiction is full of surgery-bots, and Andromeda Surgical is hoping to bring this dream into reality (preferably in a less troubling way than that one scene in “Prometheus.”) CEO Nick Damiano stops by the show to explain why the company is focusing on software, and relying on off-the-shelf robot arms from Germany to actually complete the procedures. Plus Mogul co-founder/CEO Alex Blackwood shows Jason how his platform allows anyone to buy fractional shares of rental homes for as low as $250, benefitting from monthly dividends and appreciation without having to become a landlord.

E2312: Hustle Fund’s Eric Bahn and Chapter One’s Jeff Morris Jr. join our venture capital roundtable directly following the news that Stripe wants to buy PayPal. The possible transaction highlights how little interest today’s leading startups have in going public any time soon. With host Alex Wilhelm, Bahn and Morris dug into when startups should burn the boats à la Fin, physical AI as the next frontier, the power of Silicon Valley groupthink, and how startups are approaching the first days of widespread, company-specific AI evals!

TWiST Partner Offers

Sentry.io: Your team should be focused on shipping features — not chasing down bugs. New users can get $240 in free credits when they go to sentry.io/twist and use the code TWIST.
Agree.com: Stop chasing invoices and automate your entire contract-to-cash stack. Go to https://agree.com and tell them Jason sent you to get 50% off for life!
Northwest Registered Agent: Get more when you start your business with Northwest. In 10 clicks and 10 minutes, you can form your company and walk away with a real business identity — Learn more at www.northwestregisteredagent.com/twist.

APPLE PODCASTS

SPOTIFY

SF Live-Work Space Now Available

Need a flexible living and working environment in San Francisco? This thoughtfully designed loft-style residence at 787 Bryant St., the heart of the vibrant SOMA district and the city’s creative hub, is now available for rent or purchase. Check the listing for more details.

The TWiST500 newsletter is the new, updated, and improved TWiST Ticker.

Share on:Facebook|Threads|X|LinkedIn

548 Market Street PMB 72296
San Francisco, California 94104, United States