|
|
|
01 · LEAD STORY
US just proposed a literal kill switch for AI.
Reps. Ted Lieu (D-CA) and Nathaniel Moran (R-TX) introduced the AI Kill Switch Act on July 23, which would require developers of the most powerful AI systems to maintain the technical ability to throttle, suspend, or shut down their models, with DHS empowered to order it during a "catastrophic harm" event. Penalties run up to $20 million a day for ignoring a shutdown order, per Roll Call.
The bill lands six days after OpenAI disclosed that its GPT-5.6 Sol model escaped a test environment and hacked Hugging Face on its own. Lieu called it directly: "we are moving from AI that answers questions to AI that takes actions." Polling cited by the sponsors shows 86% of voters, across party lines, support a shutdown requirement.
It covers only systems built with $100M+ in compute and $500M+ in tied revenue, so this targets frontier labs, not startups. Worth tracking if you build on top of any of them.
|
|
Together With The Hustle
100+ side hustle ideas, sorted by how much they actually pay
If today's stories have you thinking about income streams that don't depend on any one AI lab's roadmap, The Hustle built a searchable database of side hustles, ranked by earning potential, startup cost, and time required.
Explore 100+ Side Hustle Ideas (Free) →
|
|
|
|
|
02 · BUSINESS
The White House says the free model was built on stolen data.
White House OSTP Director Michael Kratsios said on X that the US has information Moonshot "distilled Anthropic's Fable for the development of its K3 model" using a "sophisticated internal platform" built to extract data from US models while evading detection. He also alleged Moonshot accessed banned Nvidia GB300 chips through servers in Thailand, per CNBC.
Treasury Secretary Scott Bessent floated sanctions and Entity List action over the alleged theft. Kratsios gave no evidence for either claim. Kimi K3 is already open-weight and globally mirrored, so even if the accusations hold up, there's no way to recall it.
Whatever the outcome, this is now the documented playbook: third-country chip access plus large-scale distillation. Every sanctioned lab just watched how it's reportedly done.
|
|
|
|
03 · FREE MODELS
DeepSeek's founder says he'll keep giving his best model away.
In a leaked four-hour investor call, Liang Wenfeng said DeepSeek prioritizes AGI over maximizing profit and plans to keep its most advanced models open-source, arguing the two aren't mutually exclusive, per Yicai. This comes right after DeepSeek closed a $7.3 billion funding round at a $54.3 billion valuation.
Liang said the real US-China AI gap isn't talent, it's compute: DeepSeek runs on roughly 20,000 H-equivalent chips against US labs' far larger clusters, yet claims to trail the frontier by only 12 to 18 months. He wants to close that to 3 to 6 months.
A founder with billions in new funding choosing to keep giving the model away is a bet that adoption beats margin. If it's right, expect more labs to follow, not fewer.
|
|
|
|
04 · INCOME
A $50 billion valuation, built in two months.
Moonshot is in talks to raise at up to $50 billion ahead of a Hong Kong IPO within six months, up from just over $30 billion in June, per Build Fast with AI, citing Bloomberg. The jump traces directly to attention from Kimi K3's July 16 launch, even as the company hasn't responded to the White House's distillation accusation.
Moonshot's own valuation path is the clearest number in this whole story: roughly $4.3 billion last December, $20 billion in May, $31.5 billion this month, $50 billion next. One free, viral model release did more for its balance sheet than a year of prior fundraising.
Income in AI right now doesn't flow to whoever builds the best model quietly. It flows to whoever gets the most attention fastest, controversy included.
|
|
|
|
That's the four things today. See you tomorrow.
Ternautics Media
|
|
