Instagram | Sponsor this newsletter

My fellow AI explorers

Elon's building a phone (he says he isn't), Alex Karp is furious about tokens, and a Stanford-MIT coalition just tried to fix the mess that is AI flaw reporting. Let's get into it.

In today’s edition:

• 📱 SpaceX's mystery AI "handset," and why Musk wants to beat OpenAI to your pocket

• 🗣️ Palantir's Karp declares war on "tokenmaxxing"

• 🚩 The 30-second play: how to actually report a broken AI model

• 📰 Claude Science, Claude’s Sonnet 5, Fed chair Warsh on AI inflation, Cisco's 90k AI agents, and Together AI's $800M raise

Six people doing the work. Your headcount is one.

Your finance close runs in #finance. Stripe and QuickBooks reconciled, runway updated, posted Sunday night without you asking.

Engineering review lands in #eng. Viktor pulled the open PRs, left comments on auth-refactor, flagged a dependency blocking api-pagination.

Campaign brief lands in #growth: Meta CPA up 18%, recommendation to pause broad match, a draft landing page already deployed for the variant test.

You hired him on day zero. He lives in Slack and Microsoft Teams alongside your contractors and investors, connects to 3,000+ tools, pushes back when you ship something dumb.

"Viktor is now an integral team member, and after weeks of use we still feel we haven't uncovered the full potential." Patrick, Director, Yarra Web.

Get Started for Free

Elon Musk

AI's Next Device Might Not Come From OpenAI After All

Before it went public, SpaceX reportedly showed investors a prototype of a "handset-like" AI device. It’s slimmer than an iPhone, according to the Wall Street Journal.

Musk, naturally, is calling the whole report "utterly false." That denial is doing a lot of work because everything else about the timing lines up.

• 🛰️ The device is reportedly built on a proprietary OS and wired directly into xAI, the Musk AI company SpaceX absorbed earlier this year… meaning no Android, no Apple, no middleman

• 📶 SpaceX has been signaling wireless ambitions for a while now, with Starlink Mobile positioned as a real Verizon/AT&T competitor. One analyst has even floated the idea of SpaceX acquiring T-Mobile outright

• 🥽 This is happening while OpenAI's own hardware push, built with Apple's former chief design officer Jony Ive, reportedly keeps stumbling, prompting the company to poach another Apple exec (Vision Pro lead Paul Meade) just last week

Here's the read: this isn't really about phones. It's about who owns the interface layer once chat-based AI stops living inside apps.

Musk watching Altman struggle with hardware and deciding "I can do that, but better" is very on-brand. But the AI device graveyard (RIP Humane, RIP Rabbit) is a reminder that wanting to ship a gadget and having people want to buy it are two very different problems.

Palantir

Palantir's Karp Just Called Token Pricing a National Security Issue

Alex Karp went on CNBC's Squawk Box this week and did not hold back: "Something has gone completely wrong, and the basic view among enterprises in this country is ‘I'm going to chillax and waste my time with tokens, I'm gonna get no value, and they're gonna get my IP.’" He insists it's not shade at Anthropic's Dario Amodei or OpenAI's Sam Altman. It's "reporting." Sure, Alex.

• 💸 Karp's term for it is "tokenmaxxing": enterprises burning cash on per-token billing with no clear return

• 🤝 The timing isn't subtle: this comes right after Palantir and Nvidia announced a Sovereign AI system built on Nvidia's open Nemotron models, letting government agencies and enterprises own their model weights instead of renting intelligence by the token

• ⚔️ Karp framed it as a defense issue too, questioning whether "warfighters" should trust their AI stack to Silicon Valley consensus at all

The subtext is obvious once you notice it. Karp's criticism of the token model doubles as a pitch for the model-agnostic, own-your-weights product Palantir just shipped with Nvidia.

That doesn't make him wrong about enterprise frustration: Uber and Microsoft have already capped employee access to pricier AI coding tools after budgets blew out. But it's worth reading his "reporting" as a competitor's sales pitch wearing a lab coat.

30-Second AI Play

How to Actually Report a Broken AI Model

Right now, if you find a serious flaw in a deployed AI system, there's no clear place to send it. Researchers duplicate work across a dozen fragmented forms, and the people who receive reports rarely share them with anyone else. A team of researchers just proposed a fix called FLARE-AI, and it's worth understanding in case you ever need to use it.

1. Classify the flaw first. Decide whether it's a safety issue, security vulnerability, bias/fairness problem, or a strict-liability case (like CSAM-adjacent risks). This determines which fields the form asks for next

2. Document it in the standard format. FLARE-AI's structure was built with feedback from 49 experts across 32 organizations, so it's designed to be triage-ready instead of a wall of unstructured text

3. Let the system route it. Rather than you guessing which of 12+ different reporting systems to use, FLARE-AI is built for interoperability: one report, multiple relevant recipients

4. Expect it to take about 10 minutes if you're using the reference implementation, versus the redundant, multi-form slog researchers describe dealing with today

If you do red-teaming, dogfood frontier models, or just stumble on something a model shouldn't be doing, this is the closest thing the ecosystem has to a "report it once, reach everyone who needs to know" button.

Advertise to 174k engineers, CTOs, and more choosing what tools their companies build with.

👉️ Learn how to sponsor this newsletter

Other Relevant AI News!

🔬 Anthropic launched Claude Science, an AI workbench that plugs into 60+ scientific databases and runs on your own lab hardware, cutting one research team's two-year literature review down to weeks.

🚀 Anthropic launched Claude Sonnet 5 as the new default model for Free and Pro users. It boasts near-Opus agentic performance at a fraction of the price, priced at $2/$10 per million tokens through August 31.

📈 Fed Chair Kevin Warsh weighed in on AI and inflation, saying the AI-driven capex boom should eventually show up on the supply side, even as near-term inflation stays too hot for comfort.

🏢 Cisco is rolling out AI agents to all ~90,000 of its employees, with AI now drafting 80–90% of its first-pass financial filings.

💰 Together AI raised $800M at an $8.3B valuation, betting on enterprises fleeing pricey frontier models for open-weight alternatives. Ironically, it’s the same frustration Karp aired above.

Golden Nuggets

• 🧩 The device race (SpaceX, OpenAI/Ive) and the token-pricing backlash (Karp, Together AI) are actually the same story from two ends. Labs are racing to own the interface and the infrastructure, while everyone downstream is trying to avoid getting locked into either

• 📋 Bookmark FLARE-AI if you ever need to report a model flaw: it's the first system actually designed to be used by non-security-researchers too

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