Proof of location. One call returns a hardware-signed yes or no about whether a device is inside a region. Verifiable offline. Free to 1,000 proofs/month.

London
Location is the substrate for human verification: only a real person can hold a phone and be in a place. At our first startup we built GPS-free geolocation for phones by fusing inertial and radio signals, runs with NATO special forces. @OctetProof lets anyone prove location that way, without trusting GPS or any single sensor on the device.
This is a big enough problem that there's not just one startup idea here, but several. For example, there will probably be one variant meant to filter out agents, but another meant to welcome them in an orderly way. This also seems an idea you could build more on.
1
4
125
Utah passed a law that requires adult sites to know whether a user is physically in Utah. VPN or not. Pornhub's parent company, Aylo, filed a lawsuit saying that the law asks the impossible. That there is no way to really know where a device is. But that isn't true. A VPN moves your traffic. It doesn't move you. Age verification is a location problem at its core. If regulated companies want to avoid verifying the age of their entire user base, they will find a way to move on from broken and antiquated geofencing practices. Our comment ↓
273
Octet retweeted
agree on OLV. telemetry helps prove performance, but residual value still depends on a (future) buyer. that makes the more interesting financing object the work itself. a robot under contract is a cash flow. missing layer is a unit of account for output (eg, picks within a spec, uptime in a bounded environment, deliveries inside a geofence, ht @OctetProof) with risk priced as variance against the contract. insurance likely comes first becuse it can underwrite that variance before a real secondary market exists. inspect the release, price the lost distribution, build premium history. all that becomes credit file.
Replying to @antavedissian
6/ Why not just lend against the robots themselves? Lenders price equipment by its orderly liquidation value: what they'd get if they repossessed it. A Toyota forklift has a known one. A new robot brand with no used market mostly doesn't. @jimbelosic explains it well 👇 nitter.net/jimbelosic/status/2045…
1
1
5
474
Where the record meets the physical.
Two words for this series. Fact. Facticity. A fact is something asserted. Someone produces it and puts it into circulation. It has an author. People love facts - because they make them! Facticity is Sartre's word for the situation you are already in: your body, your birth, your time, that you are somewhere at all. It is as broad as the whole of what you did not choose. You are free in what you make of it, never free of having it. Facticity changes just like facts. The difference is that it is not authored. It is given. It is. Facts are made. Facticity is simply there.
4
239
Gambling doesn't happen in the ether. Fraud isn't instigated in a fiber optic cable. A user is not ones and zeros. @drewarenth, COO of Octet, on binding digital decisions to a real place - and why location is the final primitive.
I'm a farmer and I build cybersecurity technology. This piece is about why those are not two separate jobs. The digital and physical have advanced on separate vectors for decades. Physical AI is where they are (meant to be) coming back in sync.
Article

Bon à Tirer

Humans and technology are losing their reference proof We need to find our way home. I am a techno-optimist. My first company brought the first AI products to global development and defense. I am

1
2
233
Octet retweeted
I'm a farmer and I build cybersecurity technology. This piece is about why those are not two separate jobs. The digital and physical have advanced on separate vectors for decades. Physical AI is where they are (meant to be) coming back in sync.
Article

Bon à Tirer

Humans and technology are losing their reference proof We need to find our way home. I am a techno-optimist. My first company brought the first AI products to global development and defense. I am

1
2
7
38,285
We looked up the "location" of an IP address to write this post. Used four different geolocation services. DB-IP said Toronto. IP2Location and ipwho.is said San Francisco. MaxMind returned "United States" with no city at all, and a very precise-looking coordinate that appears to be a field in Kansas. The furthest two are 3,644 km apart. This kind of divergence is the norm: IP databases disagree on the same address by 620 km on average. The IP address belongs to @Cloudflare. Since August 27 they have had an incident open: the commercial databases that turn IP addresses into location are returning the wrong country for WARP users. Cloudflare can't fix it, because it doesn't own those databases. That's MaxMind, IP2Location, Digital Element. Not one of them can see where you are. They infer it. Cloudflare's documentation is direct: "IP geolocation is an estimate, not an exact science. There is nothing that inherently binds an IP address to a physical location or country." Payment platforms, exchanges and licensed markets use that estimate as the signal. A user opens your app from a jurisdiction you're not licensed to serve. The lookup places them somewhere you are. You let them trade. Months later a regulator asks where that user actually was, and your evidence is a database reading that has since been corrected. Sanctions breaches are strict liability. A phone can settle it directly. Is this device inside this region, right now. Signed by a key that never leaves the chip. Verifiable offline. No coordinates leave the device. The location check doesn't have to be a guess.
1
5
498
Will be going on @MTSlive soon to talk about the provekit.org v1 announcement and how zero knowledge cryptography will reshape privacy, digital identity and World in the age of AI.
We are live and covering some big stories today: - openai's recurrent depth & looped transformer design decisions for coming astra models, as well as the discourse that's followed from this - the release of gemini 3.8 flash, as we take a closer look at the model's capabilities - elon teasing grok 4.7's release on september 11th - us government siding with openai against the new york times Following this up with some fun interviews: - @kenbwork & arjun banerjee of latch bio, talking about their latest work with @SpaceXAI on Grok 4.6's biological capabilities - @ShanuMathew93 to discuss some of the recent data center pushback, commentary from earnings reports, and some of the recent compute deals happening at frontier labs and neoclouds - @dcbuilder research engineer @worldcoinfnd to talk more about what he's working on, the state of ZK proofs, and the real world utility of this technology - @tgillibrandny founder of american perpetuals to talk about his experience trying to bringing to perps to america, and why it matters - @robertwrighter author of Nonzero and The God Test to tell us more about his new book and the power / narrative behind this ai revolution unfolding in front of us - @mohitaron founder and ceo of SciFin on the $44m seed round and what he's most excited about with this new journey - @AndreyFradkin professor @BUQuestrom, discussing his thoughts on the Cosean Singularity and how he's defining it - @TheRealNamzoo founder and ceo of airbound, on the company's recent $37m series A, the future of flight, and how he's working to push the limits of this technology - Laurel Prime founder of Astravia to talk more about what motivated her to take the leap and become a founder, as well as what Astravia is doing to redefine travel with the help of AI
4
8
49
11,991
Better doesn't mean blocking more people. It means letting the right ones through. Our best users hit a VPN block and go to a competitor. — Ecosystem lead, L2 network
1
1
4
180
My @OctetProof co-founder Suvrit Sra just released "GPT, the Counterexample Machine". 15+ counterexamples to publicly posed conjectures, found by AI over 12 months. One killed a conjecture he posted on MathOverflow in 2017!
2
2
3
234,194
Octet retweeted
Proofs remove the need for "trust me." In fact, proofs remove the need for "me" at all. The work stands on its own or fails on its own, and either way it can be checked independently. The world is full of opinions. When it matters, we don't need opinions. We need proofs. @benjamintfels Fels elegantly summarizes @optiML work here. I'm lucky to be building simple products alongside complex minds and technologies at @OctetProof
My @OctetProof co-founder Suvrit Sra just released "GPT, the Counterexample Machine". 15+ counterexamples to publicly posed conjectures, found by AI over 12 months. One killed a conjecture he posted on MathOverflow in 2017!
3
32
173,661
Octet retweeted
I couldn't possibly think of a better reason to integration real world proof of location. Immutable - cryptographic - hardware attested - locally derived. I want @HyperliquidX in the USA and we can help. @OctetProof
North Korea's state hackers just laundered over $30 MILLION through Hyperliquid, right as the platform is trying to enter the US The Lazarus Group sold Bitcoin on Hyperliquid, flipped it into Ethereum and Solana, then scattered it across chains and exchanges to cover the trail These are the same wallets tied to tens of MILLIONS in stolen crypto, flagged back in 2024 The timing is brutal with Trump's team actively working to bring Hyperliquid into regulated US markets while this is happening on it Nothing invites regulators faster than a sanctioned government washing money through the platform you're trying to legitimize...
2
4
229
Your "US persons only" control is probably a sentence in a policy document.
1
1
3
220
Ironvale is invented. The mechanism is not: under ITAR, giving a foreign national access to controlled technical data is an export, whether or not anything ships.
42
Someone on your payroll is probably not where your records say they are.
1
4
200
This one isn't a forecast. Five people pleaded guilty in 2025 to hosting company-issued laptops in US homes so North Korean workers appeared local — 136 US companies affected, $2.2M generated for the regime. DOJ has charged more than 40 people across these schemes.
2
38
Su geobloqueo probablemente se ejecutó una vez, al registrarse. Compruebe cuándo fue la última.
1
3
160
Este no es un pronóstico. La OFAC llegó a un acuerdo con Kraken por 362.158,70 dólares por 826 operaciones de usuarios que parecían estar en Irán. El factor agravante: los controles de geolocalización se aplicaron al registrarse, pero no a la actividad posterior. La exposición máxima era de 272 millones.
1
32
Your geofence probably ran once, at signup. Check when it last ran.
1
1
5
163
This one isn't a forecast. OFAC settled with Kraken for $362,158.70 over 826 transactions by users who appeared to be in Iran. The aggravating factor: geolocation controls applied at onboarding but not to later transactional activity. Maximum exposure was $272M. ofac.treasury.gov/media/9295…
3
45