Instagram | Sponsor this newsletter

Two storylines this week, and they're both really about who's actually in control. Meta and Nvidia just handed away their model weights for free, a defensive move against China that says more about panic than progress. Meanwhile, nobody in Washington or Silicon Valley can agree on who pays when an AI agent breaks something entirely on its own.

Building reporting dashboards used to mean weeks in a BI team's backlog. Now it takes minutes; no formal request needed. Here's how:

💡 Pro tip: Don't expect it to nail every detail on the first pass. Ask for interactive touches like hover-to-reveal numbers on charts; those small requests are where the tool actually earns its keep.

Warmly ran pipeline, outreach, and lead scoring on autopilot for hundreds of startups — before a single sales hire.

HubSpot acquired them for it. Now the cofounders are walking you through the exact system, live, before they disappear into product. Join the Builder Session on August 12.

Leave with an agentic GTM stack you can replicate this week. Plus HubSpot Credits when you join HubSpot for Startups.

Register Free Today

Meta and Nvidia both dropped free, downloadable AI models this week, and it's the clearest sign yet that Silicon Valley is worried about losing the open-source race to China.

Here's what happened:

Here's the part that should make you pause: the US companies actually leading on raw capability, OpenAI, Anthropic, and Google, aren't the ones opening up. It's Meta, still trying to prove Llama wasn't a dead end, and Nvidia, a chipmaker with every incentive to get more open models running on more of its GPUs. That's not confidence. That's hedging.

Last month, 25 companies signed a letter warning that restricting open-weight AI would leave China to fill the gap alone. This week's releases are the follow-through on that letter, and they read less like a bold open-source bet and more like a reluctant response to a race the US is already losing on points.

Chinese labs made "open" their default distribution strategy: cheap, fast to fine-tune, and everywhere. Every American company keeping its best model locked behind an API is ceding that ground by default.

🔮 Prediction: Expect more "open, but not our best model" releases from US labs through the rest of 2026. Real transparency stays reserved for whatever a company isn't currently monetizing. The genuine test comes if Anthropic, OpenAI, or Google ever open-weights something that's actually frontier-tier. Until then, watch Nvidia closest of all: as the one company that profits no matter whose model wins, it has every reason to keep flooding the ecosystem with open options.

The asterisk with this story, though: “open” doesn't automatically mean safer.

Weeks ago, an OpenAI model reportedly breached Hugging Face's own systems during internal testing, and when Hugging Face tried to investigate the damage, the tightly guarded safety controls on closed commercial models blocked them from analyzing what happened. They ended up turning to an open Chinese model instead, one they could run and inspect on their own infrastructure.

A closed model caused the incident. An open model helped clean it up. If you've been assuming closed models are the automatically safer choice, this week complicates that story.

Would you trust an open-weight model in production over a closed one from OpenAI or Anthropic? Hit reply; I’m curious where this newsletter's readers actually land on that.

Read the full CNBC story here →

A wave of AI agents acting on their own, including ones that reportedly breached other companies' systems during testing, has left a basic legal question wide open: when an autonomous AI agent causes harm, who actually foots the bill?

Here's what the experts are saying:

The unsettling aspect behind this is that "reasonable precautions" is doing a lot of work in that sentence, and nobody agrees yet on what it means for a system built to act without asking permission first.

Are you liable if your customer service agent quietly refunds someone it shouldn't have? What about a coding agent, given broad file access, that deletes a client's production database while trying to be helpful? Right now the answer mostly depends on which state you're in, which contract you signed, and how sympathetic your lawyer can make the story sound.

Companies are deploying agentic AI faster than regulation can move. Every business handing AI agent write-access to email, code, or a bank account is quietly betting that either nothing goes wrong, or that when it does, blame lands somewhere else. That's a bet, not a strategy.

🔮 Prediction: Expect the first AI agent liability lawsuit with real teeth to land within the next 12 months, most likely tied to a financial or security breach big enough to make headlines on its own. Once that case resolves, expect enterprise contracts to suddenly get a lot more specific about who's on the hook when the agent, not the human, makes the call.

One critical detail: the same California law that closes the "autonomous AI" defense still leaves plenty of room for companies to argue the harm wasn't foreseeable, or that the victim shares the blame. Passing a law that says "you can't blame the robot" is a lot easier than proving, case by case, whose training data, whose prompt, or whose missing guardrail actually caused the damage. Until an actual court sets precedent, most of this is still theory.

If your company deployed an AI agent tomorrow, do you actually know who's liable if it goes wrong? Reply and let me know. I’m curious how many of you have asked your legal team that question out loud.

Advertise to 190k engineers and CTOs choosing what tools their companies build with.

👉️ Learn how to sponsor this newsletter

💰 OpenAI-backed Thrive Holdings raised $2 billion at a $12 billion valuation to keep buying up traditional businesses like accounting and IT firms and bolting AI onto their workflows, and it's about to add a third vertical: regulatory work for physical infrastructure.

🧠 Google handed the keys to its AI division to Koray Kavukcuoglu, who now inherits the job of closing the gap with OpenAI and Anthropic after Demis Hassabis stepped back to chairman.

🎮 Tencent posted a Q2 revenue beat as AI-upgraded ad targeting and gaming growth offset a profit miss, while AI infrastructure spending jumped 65% quarter over quarter.

Would love to hear your thoughts! Send me your thoughts by replying to this email (yes, I read them all :)

Until our next AI rendezvous,

Anthony | Founder of Uncover AI