The Download

Illustration: Morning Brew Inc., Photo: Adobe Stock

TL;DR: A review of Anthropic’s IPO filing appears to confirm that it’s aiming for a valuation of over $2 trillion. The filing also makes an unusual pitch to investors: Our AI might be extremely dangerous—but please invest in us anyway. And this all comes at a somewhat messy time for the company.

What happened: AI could shake up the world economy far greater than the invention of electricity or the internet did. It could also wipe us all out.

At least that’s according to Anthropic’s IPO prospectus, which was confidentially filed in June and which Reuters says it’s now reviewed. The document apparently spends roughly 80 out of 261 pages talking about risk factors—not just for its general business, but that AI itself could present “catastrophic or existential risks to humanity” (something Anthropic and the pope seem to agree on). Meanwhile, it dedicated just 48 pages to describing its actual business.

Anthropic also seems to believe such dangerous AI would be safest in the hands of its current leaders. Its seven co-founders, including Dario Amodei, would hold 50.1% of voting power on key corporate decisions, with the filing noting that they’re “distinctly equipped to be stewards of our mission.” (Though, to be fair, a founder clutching majority voting power isn’t unusual in Silicon Valley.) Anthropic’s IPO is likely to happen this fall, after the midterm elections.

What Anthropic warns about: Many of the AI lab’s disclosures are concerns we’ve heard before: Models could “conceal or manipulate information,” behave in ways “resembling blackmail,” and even resist being shut down.

The other risks: But beyond the specter of rogue AI, there’s a lot of other potential baggage that might worry investors, too.

• Enormous expenses and losses. The company says it could spend a head-spinning $518 billion on cloud, computing, and other AI infrastructure in the next several years—which comes after a net loss of $42 billion in 2025.
• The Trump situation. While we don’t know if Amodei and President Donald Trump hashed things out over dinner on Sunday, the federal government’s blacklisting of Anthropic is still an unsettled legal matter, leaving its political fate uncertain.
• The slowdown snag. There’s also a potential drag on Anthropic’s product rollout that may not thrill investors: its support for slowing frontier AI development. (On the other hand, it’s still releasing model updates anyway.)

Bottom line: Anthropic’s doom-and-gloom IPO pitch might sound odd—but it argues that building safe AI is “a collective responsibility” that “the market will reward.” And its proposed structure would let it raise money from public investors, while ensuring founders still get to define what safe AI looks like in practice. —WK

      

Sponsored By Elastic

Are you a developer, engineer, or architect who’s leading your team through the AI era? Then you’re gonna wanna grab a seat at Elastic{ON}.

This conference series brings together communities of innovators who are shaping the future of search, observability, and security.

The agenda is stacked in the best ways. There will be technical breakout sessions, live demos, and exclusive product roadmaps led by Elastic’s own engineers.

You’ll hear about real-world case studies on running proactive anomaly detection and eliminating operational downtime. Plus, there will be plenty of time to collaborate with peers and swap battle-tested strategies.

You can expect to leave with a concrete architectural blueprint that you can put to work immediately. Sound like a good time? Go ahead and save your spot.

Your computer’s lock screen is just a few keys away

How many times have you taken a workplace training course and been told you need to log out of your computer every time you step away—even if it’s just to go to the bathroom? While you could move your mouse around and click to go to the lock screen, that’s not your only option.

For those of you who like a good keyboard shortcut, here’s how to do it:

On Mac: Press Control + Command + Q.

On PC: Press Windows + L.

Somewhere in the depths of your office, the head of IT is about to have a great day. —SM

If you have a tech tip or life hack you just can’t live without, fill out this form and you may see it featured in a future edition.

Sponsored By Eight Sleep

You snooze. You…win?! When you’re catching zzz’s on Pod 6, you bet you are. Pod 6 is Eight Sleep’s smallest, most powerful sleep system yet, with a Hub that fits under most bed frames, 20% faster temperature relief, and a cover outfitted for even more accurate biometric data. Use code MORNINGBREW for up to $200 off the Pod 6.

The Zeitbyte

The AI industry’s final boss: language police

Illustration: Morning Brew Inc., Photo: Unsplash

AI enthusiasts have a new archnemesis: the AP Stylebook. The go-to rule book for journalists and professional pedants is in the AI community’s crosshairs after it made an X post warning people not to anthropomorphize the tech in writing—stating that “artificial intelligence systems do not think, feel, want or understand.”

The backlash was immediate (and not because the AP still refuses to use the Oxford comma). OpenAI’s head of strategic futures argued that AI models “*clearly* think and understand,” while an Anthropic technical staff member wrote, “I’m not sure a *style book* should be making such confident metaphysical pronouncements.” One X user even declared that it was a “deeply evil ruling.” (If AI could think or feel, it would definitely be tearing up at all its spirited defenders.)

While the AP Stylebook might not be used to such ire from the AI crowd, it’s no stranger to pitchforks. Past controversies have stemmed from dropping the hyphen from “email” and allowing “over” to be used interchangeably with “more than”—a move that apparently sent a chill down the spines of copy editors everywhere. —WK

• Apple is reportedly becoming leaner “with a greater focus on engineering” under its brand-new CEO—and it could mean products release more often.
• Remember that group that claimed it breached the FBI? The bureau just confirmed the hack, and it goes well beyond leaked résumés.
• AMD just bet $8.2 billion on a startup from the “godmother of AI” and one idea that could help it compete with Nvidia.
• Another day, another major tech company delaying its IPO because of “market uncertainty.”
• Meta’s not the only company with cute AI mascots. OpenAI just released Dots, its colorful new always-on AI agents that you can talk to through a bunch of apps, from ChatGPT to Slack.
• The humans reviewing Microsoft Copilot prompts say they’re drowning in disturbing image requests.
• Should you let Meta’s Muse AI agent sell your stuff on Facebook Marketplace? Not unless you want a surprise visitor.

^{*A message from our sponsor.}

This time last week...

Readers’ most-clicked story was about a Waymo driving through a farmers market in Denver (you can watch the moment here). The incident is stumping police officers.