Your Terms of Service Are a Product Decision
Instinct hit a $2.5 billion valuation while still invite-only. It also spent the same week defending a perpetual training license buried in its terms. For first-time founders, that sequence is the lesson.

What happened to Instinct in one week?
Instinct, a consumer AI assistant run by Spear Street Technology and led by 23-year-old founder Noah Shinn, closed a $250 million Series B co-led by Index Ventures and Benchmark. Total raised: $350 million. Valuation: $2.5 billion. The product is still invite-only. It has never had a public launch.
The round was reported on August 26 and 27. Two days earlier, on August 24, TechCrunch published a piece about the same product with a different headline: privacy and security concerns.
Same product. Same week. Two stories.
Early testers loved what it did. You connect it to iMessage, WhatsApp, your email, your calendar, and then you text it like a person. Book this. Cancel that. Find me an apartment. Shinn told the Wall Street Journal some users had bought homes through it. That's not a demo. That's a real agent doing real transactions.
Then people read the terms.
What were the actual complaints?
Four things, and they're worth listing individually because each one is a different kind of mistake.
The terms of service granted a broad, perpetual license to use user materials for training. Perpetual. Not "during your subscription." Forever.
The assistant could capture screen recordings and keyboard inputs. Necessary for the product to work, arguably. Also the exact description of a keylogger if you write it without context.
Users reported that it retained and summarized emails after access was revoked. Revoking access is supposed to mean something. When it doesn't, that's not a policy question, that's a broken promise.
And one user said it sent an email on their behalf without asking first. An agent that can act is an agent that can act wrongly.
Security researchers also found it could be phished. Which follows: if a thing reads your inbox and takes instructions from text, then anyone who can put text there can try to give it orders.
Through all of it, the team stayed quiet on X.
Why does this matter to a founder with 40 users?
Because the gap between "our product needs this permission" and "our terms say we own this forever" is where most consumer trust gets destroyed, and it almost never gets destroyed on purpose.
Here's how it usually happens. You're at seed stage. You need a terms of service to launch, so you copy one from a similar company or pay a lawyer $2,000 for a template. The lawyer's job is to protect you from liability, so the document grabs the widest rights it can.
Nobody at your company reads it as a product document. It sits there for eighteen months. Then you get popular, someone screenshots clause 4.2, and now you're the company that claims a perpetual license to people's private messages.
You didn't decide that. But you shipped it.
The fix costs nothing at seed stage and gets expensive later. Read your own terms once, out loud, as a user would. Every place the document takes more than the product needs, cut it.
What's the difference between permissions you need and permissions you take?
A useful test: for every permission your product requests, write one sentence explaining what breaks if you don't have it. If you can't write that sentence, you don't need the permission.
Instinct needs inbox access. Obviously. You can't have an assistant that books flights without letting it see the confirmation emails. That's the deal, and users who signed up understood it.
But a perpetual training license on the contents of that inbox is a separate ask, bundled into the same yes. So is screen recording. So is retention after revocation.
That bundling is the thing users react to. Not the access itself. People hand Gmail their entire correspondence history and don't think twice. What they object to is finding out that the yes they gave covered more than they thought.
So unbundle. Separate consent for separate things. Training on my data should be its own checkbox, defaulted off. Screen capture should be a mode I turn on, not a condition of installing. The product gets slightly worse. The trust curve gets much better.
One more piece, less obvious. Revocation has to actually revoke. If a user disconnects their email and your system still holds summaries, you need to say that in plain words before they disconnect, not after they discover it in a support thread.
Your AI co-founder is ready when you are.
Foundra turns everything in this article into an actual plan. Validation, customers, pricing, launch. In one place, in your voice, in an afternoon.
Get started→$39/month. Cancel anytime.
Does raising at $2.5 billion mean they got it right?
No, and conflating the two is a trap first-time founders fall into constantly.
A valuation is a bet on a future, priced by people who are professionally comfortable with risk. Index and Benchmark are not certifying that Instinct's data practices are sound. They're pricing the probability that a personal agent category exists and that this team wins it.
Both things can be true at once: the product is remarkable, the round is real, and the trust foundation has cracks that will cost money to fix. Valuations don't audit anything.
The useful signal isn't the number. It's what the number obligates. At $2.5 billion, Instinct now has to grow into a public consumer product, which means app store review, regulator attention, enterprise buyers asking for a SOC 2, and journalists reading every clause. Invite-only bought them time. That time is over.
What should you build before you go viral?
Not much. But the short list isn't optional.
A permissions map. One page. Every piece of user data you touch, why you touch it, how long you keep it, and what a delete request actually deletes. If you can't fill this out, your engineers are making privacy policy by accident.
A confirmation step for irreversible actions. Anything that sends, buys, cancels, or posts on a user's behalf needs a human yes the first several times. You can relax it later per user, per action type. You cannot un-send an email your product sent for someone.
A public incident channel. A status page, a changelog, an X account someone actually watches. Silence during a trust story reads as guilt, even when it's just a busy team.
Terms that match the product. Have someone non-legal read them. If they're surprised by anything, users will be too.
Mapping this out is planning work, not engineering work, and it's the kind of thing that falls through the cracks when everyone is heads-down shipping. If you want structure for it, Foundra's planning workspace has templates for working through product and go-to-market decisions like this before they turn into fires. A shared Notion doc works fine too. The point is that someone owns it.
How do you know if your terms are a liability?
Run this in twenty minutes.
Open your terms of service and your privacy policy. Search for the words "perpetual," "irrevocable," "worldwide," "sublicense," and "any purpose." Read each sentence they appear in.
Now ask: would I be comfortable if a reporter quoted this sentence with my company name attached and no other context? Because that's the only way it will ever be quoted.
Second pass. Find every place the document says you may do something. Change the ones you don't currently do to say you don't do them. Lawyers hate this because it narrows your options. Users love it because it's honest, and narrowing options you weren't going to use costs you nothing.
Third pass, the one people skip. Ask an engineer whether the system actually behaves the way the policy describes. Deletion in particular. A lot of "we delete your data" claims mean the row is marked deleted and the backups keep it for ninety days. Say the ninety days.
I've watched founders treat this as a compliance chore. It isn't. It's the cheapest brand asset you'll ever build, and it only works if you do it early.
Key takeaways
- Instinct raised $250 million on August 26-27, 2026 at a $2.5 billion valuation, bringing total funding to $350 million, while still invite-only.
- Two days earlier, TechCrunch reported concerns over a perpetual training license, screen and keyboard capture, email retention after revocation, and an unconsented send.
- Trust damage usually comes from bundled consent, not from the underlying access. Users accept access they understand.
- Unbundle permissions, default training off, make revocation real, and confirm irreversible actions.
- A valuation prices a future. It does not validate your data practices.
- Read your own terms searching for "perpetual," "irrevocable," and "any purpose," then cut every right the product doesn't use.
- Silence during a trust story reads as guilt. Have a channel and a person who answers.
FAQ
How much did Instinct raise and at what valuation? Instinct raised a $250 million Series B co-led by Index Ventures and Benchmark, reported August 26-27, 2026. Total funding reached $350 million at a $2.5 billion valuation, while the product remained invite-only.
What is a perpetual license in a terms of service? It means the company keeps the right to use the material you provided even after you stop using the service and delete your account. For AI products it usually covers training rights on your content, which is why users react strongly when they find it.
Do early-stage startups actually need a custom terms of service? You need one that matches what your product does. A template is fine as a starting point, but every clause granting rights broader than your actual behavior is a future headline. Editing down is cheaper than a rewrite under pressure.
How should an AI agent handle actions taken on a user's behalf? Require explicit confirmation for anything irreversible: sending, buying, canceling, posting. Let users graduate specific action types to automatic once they've approved them a few times. Never default to autonomous on first use.
Can AI assistants be phished? Yes. Any agent that reads user content and acts on instructions can be manipulated by attacker-controlled text placed in that content, often called prompt injection. Treat incoming email and messages as untrusted input, not as commands.
Is it bad to stay quiet when your product gets criticized? Usually yes. A short factual response with what you're changing and by when does more for trust than a polished statement two weeks later. Silence gets interpreted, and the interpretation is rarely charitable.
Sources
- TechCrunch: Instinct's powerful AI assistant is raising privacy and security concerns
- TechCrunch: Viral AI startup Instinct has raised $350M at a $2.5B valuation
- SiliconANGLE: Consumer-focused AI assistant startup Instinct reportedly raising $250M
- PYMNTS: Instinct Nears $2.5 Billion Valuation as AI Assistants Take Over Daily Chores
- SC Media: Instinct AI assistant faces privacy concerns amid praise
- Tech Startups: Startup Funding News Today, August 27, 2026
You just read the theory. Ready to build the thing?
Foundra is your AI co-founder. It turns an idea into a validated business plan, a go-to-market, and your first 10 customers. In an afternoon, not a semester.
$39/month. Cancel anytime. Works in 20 languages.