Welcome, humans.

Quick housekeeping note: We’ve been testing a new voice clone tool through Beehiiv (our newsletter platform) to use Grant’s voice to do our audio summaries, but we’ve got some reports that the volume is, ahem, out of whack.

Today we’re running “Will”, our old voice. Here’s yesterday’s with Grant’s voice for comparison. Can y’all let us know in the feedback at the bottom of this email which you prefer? TY!

Now: the big news of today is Google’s new version of Gemini 2.5 Pro, which will be the one that rolls out for “enterprise-scale applications” over the next few weeks.

As with all Google models, you can test it out (for free) here in the AI Studio.

Since our main story today is all about how AI companies will probably hold onto your data, forever (yes, even your business data), it’s worth remembering that anything you put in the AI Studio can and will be used to train Google’s models.

That’s the cost of “free”, of course… YOU are the product.

Here’s what you need to know about AI today:

• The fight between OpenAI and NYT will determine your AI data privacy.

• Anthropic launched Claude Gov models for U.S. national security customers.

• Perplexity hit 780M searches last month.

• OpenAI stopped 10 malicious operations, 4 likely from China.

Your AI Chats Were Never Private (And Now We Have Proof)

Remember when you thought your deleted ChatGPT conversations actually disappeared? Yeah, about that...

A federal judge just ordered OpenAI to preserve ALL user conversations indefinitely. That includes the ones you deleted, those “temporary” chats that were supposed to vanish, sharing business secrets through OpenAI's API, everything. 

The order is part of the ongoing lawsuit from The New York Times, and it affects hundreds of millions of users.

OpenAI isn't taking this sitting down. In a blog post, they called the demand a “sweeping and unnecessary” overreach that “fundamentally conflicts with the privacy commitments we have made to our users.” Sam Altman went even further:

He also floated the idea of “AI privilege”, the concept that talking to an AI should be as confidential as talking to your doctor, therapist, or lawyer.

Not a bad idea, given the fact many people are using ChatGPT, as, well, their doctors, therapists, and lawyers…

So, who’s impacted? According to OpenAI, the order affects ChatGPT Free, Plus, Pro, and Team users, along with standard API customers. If you’re a ChatGPT Enterprise, Edu, or an API customer with a Zero Data Retention (ZDR) agreement, your data is not included in the order.

This ruling begs the question: what kind of data are we talking about, anyway? While the court order is new, data collection isn't.

This chart showed up on Reddit this week, giving a snapshot of just how much data AI chatbots could be hoarding (based on their self-reported App Store privacy notices):

1. Meta AI takes the crown, collecting a staggering 32 out of 35 possible data types (including financial info, health data, and even “sensitive information” like your political leanings).

2. Google Gemini comes in second with 22 data types, including your precise location and contacts. No surprises there—Google gonna Google.

3. ChatGPT and Microsoft Copilot are surprisingly restrained, collecting only 10 data types each. And Elon’s Grok is one of the lowest collectors on the list.

The catch: This is based on what companies claim they collect. As some have pointed out, these labels can be finicky. ChatGPT shows location tracking on Android but not iOS. Meta probably checked every box just to cover themselves legally. And Apple's reviews can't catch server-side tracking that happens after you hit send.

Why this matters: The OpenAI court order proves that whatever these companies claim about data deletion is essentially meaningless when a judge says, “save everything.” Your late-night anxiety spirals, your business strategy sessions, your secret project prompts—it's all being preserved for potential legal review.

This sets a dangerous precedent. Every other AI company is watching this unfold, and you can bet their lawyers are thinking, “Maybe we should just keep everything, too.” Unless we get Sam’s vision of prompter-AI privilege to protect us, we could be entering an era where every conversation with your AI could become part of a permanent record.

We wrote a massive deep dive on this topic and all of its myriad implications (and we really mean ALL) on the website here.  

P.S: This is not legal advice, obviously. We know there’s a lot of lawyers in the audience (what’s up y’all!), and we’d love to hear your take. Reply in the feedback at the bottom if you’re interested and we’ll reach out!

FROM OUR PARTNERS

200+ AI Side Hustles to Start Right Now

Remember when “side hustle” just meant driving for Uber or selling crafts on Etsy? Those days are officially over.*

AI isn't just changing business, it's creating entirely new income opportunities that didn't exist six months ago. The Hustle's guide features 200+ ways to make money with AI, from (think AI content creation) to advanced ventures (custom AI consulting).

Each comes with realistic income projections and resource requirements, so you're not flying blind. Join 1.5M professionals getting daily insights on emerging tech and business opportunities. Because while everyone else is worried AI will take their job, you'll be building new income streams with it.

Get the guide

Prompt Tip of the Day

Here’s two more prompt tips that seem obvious in hindsight, but if you don’t know them, will probably REALLY help you:

1. Tell your AI exactly how to answer you. Define the desired output format or style upfront. If you need a bulleted list, a table, or a tweet-length response, include that instruction.

   1. For example: “List the output as 3 bullet points.” The model will oblige and structure its answer in the format you asked for.

2. Big request? Tackle it one step at a time. If your prompt is complex, split it into smaller tasks or explicitly ask for a step-by-step solution. ChatGPT and other AIs do better when they can focus on one thing at a time (just like us!).

   1. For instance, first ask for an outline, then delve into each point in subsequent prompts.

Bonus: you CAN break things down step by step in a single prompt. Example:

Thinking models like 2.5 Pro or ChatGPT’s o3 inherently do this “step by step” planning for you, but if you want finer control over your output (tip #1), you can write out the steps yourself.

Treats To Try.

*Anything marked with asterisks is sponsored content. Advertise in The Neuron here.

1. *Murf AI turns your text into human-like speech in 200+ voices, so you can narrate presentations or create audiobooks without recording anything yourself.

2. Adobe released Boards, which generates images from your text prompts and organizes them into moodboards and artboards for your design projects.

3. JobForAgent matches you with builders who create custom AI solutions to automate your tasks—free to try.

4. Morning Commute lets you earn medical CE credits during your drive with 30-minute expert podcasts—free to try.

5. Onda turns your podcasts into searchable notes with timestamps so you can save key moments, then find any insight instantly instead of re-listening.

6. Omakase puts a chat/voice widget on your site that answers customer questions (Japanese company, very cool—we’ve chatted w/ them before!).

7. JammyChat scans your face and gives you a playlist that matches your mood—free to try

See our top 51 AI Tools for Business here!

Around the Horn.

• Anthropic announced new Claude Gov models designed exclusively for U.S. national security customers, with enhanced capabilities for classified materials and defense contexts.

• Anysphere, maker of AI coding assistant Cursor, raised $900M at a $9.9B valuation with revenue surpassing $500M ARR.

• Perplexity hit 780M monthly queries and is building a new browser called Comet.

• OpenAI released a new threat report that claimed it stopped 10 “malicious operations” between March and June, of which 4 probably came from China, posing as commenters who convincingly impersonated actual posters (report).

FROM OUR PARTNERS

AI-Ready Web Data Infrastructure

Your AI needs fresh data to stay smart, but web scraping is a nightmare.

Bright Data handles the heavy lifting—real-time web data collection (at large scales), ready-to-use datasets, and APIs that automatically pull organized data from public websites. Whether you're training models, fine-tuning systems, or building AI agents, they've got the infrastructure you need.

Try Now

Intelligent Insights

1. This piece argues that AI actually just mirrors the operator’s skill rather than possess independent intelligence—which is why you / your organizations should invest in prompt engineering training.

2. The Darwin-Gödel Machine represents a major breakthrough in self-improving AI, suggesting our role may shift from designers to guardians of evolving systems.

3. This eye-opening account shows how AI transformed one programmer's workflow by acting as a “virtual intern” that codes while he makes coffee.

4. AI pioneer Karpathy predicts that by 2025, 99.9% of content will be optimized for AI consumption while still written for humans.

5. This MIT spinout, Themis AI, developed a platform to teach AI models to self-report their confidence levels and automatically flag unreliable outputs; this could potentially solve the hallucination problem blocking AI deployment in high-stakes applications like drug discovery and autonomous vehicles.

A Cat's Commentary.