Pinned Tweet
🚨 We trained a large language model on 2B+ card transactions and it caught 35% more fraud.
Like the labs train models on words, we can train models on transactions. To do that, the model has to read a transaction like it reads a word. So each transaction gets broken into its parts:
- Who the merchant is
- What it sells
- How much moved
- Which channel it came through
- How long since this customer last transacted
These models are pattern matching machines. Claude “predicts” the next word in a sentence whereas our trained model predicts the next transaction a customer might make.
Repeat this across billions of transactions, and the model builds a detailed sense of what normal looks like for each individual customer.
We never told the model which transactions were fraud. We don’t want it chasing fraud directly. What we want is for it to learn what ordinary spending looks like, so anything off-pattern becomes instantly recognizable.
For any given transaction, the model produces 2 things:
1. A compact numerical summary of everything the customer has done up to that point, which in machine learning is called an embedding.
2. A "surprise" score. Basically, a measure of how unexpected this particular transaction looks given the customer's history.
We trained a foundation model that discovers the language of transactions. Every person has their own dialect that they speak, and the embedding we built discovers the family of dialects that describe payments.
The holy grail here is to take the foundation model and apply it to a brand new client (whose data was not in training set). That's what we did and had amazing results:
Across every threshold we tested, adding the embeddings caught 24% to 35% more fraud. At a common operating point, where the riskiest 1% of transactions get flagged for action, the improvement was 28%.
Our AI team wrote an entire blog post about it that digs into specifics, drawbacks, and our methodology:
sardine.ai/blog/foundatio…
We are also publishing a whitepaper titled ‘Building a foundation risk model’ that I’ll talk about more here in the coming weeks. Follow for more on this process.
One important note to end on - we didn't replace our scoring model with a neural network. Our issuing model already scores every transaction using gradient-boosted decision trees, built with XGBoost, and running on 77 features we've engineered over the years. The new embeddings were simply added to it as extra features, and then we retrained the same XGBoost model on the larger set.
Again, more specifics in the blog than I can fit here.
I’m incredibly encouraged by this progress, both for @sardine and the entire fraud & fincrime fighting community.
Like the labs train models on words, we can train models on transactions. To do that, the model has to read a transaction like it reads a word. So each transaction gets broken into its parts:
- Who the merchant is
- What it sells
- How much moved
- Which channel it came through
- How long since this customer last transacted
These models are pattern matching machines. Claude “predicts” the next word in a sentence whereas our trained model predicts the next transaction a customer might make.
Repeat this across billions of transactions, and the model builds a detailed sense of what normal looks like for each individual customer.
We never told the model which transactions were fraud. We don’t want it chasing fraud directly. What we want is for it to learn what ordinary spending looks like, so anything off-pattern becomes instantly recognizable.
For any given transaction, the model produces 2 things:
1. A compact numerical summary of everything the customer has done up to that point, which in machine learning is called an embedding.
2. A "surprise" score. Basically, a measure of how unexpected this particular transaction looks given the customer's history.
We trained a foundation model that discovers the language of transactions. Every person has their own dialect that they speak, and the embedding we built discovers the family of dialects that describe payments.
The holy grail here is to take the foundation model and apply it to a brand new client (whose data was not in training set). That's what we did and had amazing results:
Across every threshold we tested, adding the embeddings caught 24% to 35% more fraud. At a common operating point, where the riskiest 1% of transactions get flagged for action, the improvement was 28%.
Our AI team wrote an entire blog post about it that digs into specifics, drawbacks, and our methodology:
sardine.ai/blog/foundatio…
We are also publishing a whitepaper titled ‘Building a foundation risk model’ that I’ll talk about more here in the coming weeks. Follow for more on this process.
One important note to end on - we didn't replace our scoring model with a neural network. Our issuing model already scores every transaction using gradient-boosted decision trees, built with XGBoost, and running on 77 features we've engineered over the years. The new embeddings were simply added to it as extra features, and then we retrained the same XGBoost model on the larger set.
Again, more specifics in the blog than I can fit here.
I’m incredibly encouraged by this progress, both for @sardine and the entire fraud & fincrime fighting community.
3
6
87
6,824
Super excited for this. Come hear about how we are fighting fraud using Jev. Co-hosting the event with our friends from @a16z. I will be on a panel together with the CTO of @typesafeai
Next Wednesday, November 4, Sardine and @typesafeai AI are cohosting Jev and the Future of Fraud Fighting in San Francisco's Financial District.
Jev, TypeSafe AI's System One model, can score risk in about a tenth of a second. @soupsranjan, our co-founder and CEO, and Erik eSafe AI, will dig into where it fits in a live fraud stack.
The two of them will share findings from real-world fraud use cases:
• Detecting transactions linked to fraud rings
• Identifying suspicious signals during onboarding
• Automating KYB risk checks
Soups is bringing results from a real fraud ring he tested, where Jev edged out well-built velocity rules and beat an LLM that cost about 25 times more.
Yoko Li, a partner at a16z, will moderate the conversation.
Risk teams weighing AI for approve-or-decline calls will get a straight read on what Jev looks like with real fraud running through it. The TypeSafe AI team has impressed us since Jev launched in September, and we're thrilled to host this one with them.
Drinks with fellow fraud and risk leaders are on us after the conversation. Grab one of the limited spots here: lnkd.in/en8hmEhG
0
0
9
486
This is wise on Walmart's part. One of the biggest gaps in agentic commerce is that user's purchase and transaction history gets bifurcated into: those made by the human and those made by agent.
This means human never gets the loyalty points from the Agentic shopping
And all existing chargeback rules like Visa Compelling evidence 3.0 go out the door - human can claim they never authorized the agent. And agent shopping will show different device id, and IP address than whats commonly associated with the human
More in our Know Your Human whitepaper coming out soon
This means human never gets the loyalty points from the Agentic shopping
And all existing chargeback rules like Visa Compelling evidence 3.0 go out the door - human can claim they never authorized the agent. And agent shopping will show different device id, and IP address than whats commonly associated with the human
More in our Know Your Human whitepaper coming out soon
Walmart’s MCP server is generally misunderstood; it doesn’t provide carte blanche catalog / transaction access and, if anything, is further support for the idea that Agentic Commerce is a Mirage.
First, Walmart is allowing third-party agents to *connect to Sparky*, its own on-platform agent. This is an agent-to-agent gateway. Walmart retains the consumer data and first-party consumer relationship. And Walmart exercises fundamental control over product discovery and recommendations if Sparky is the agent interface.
Second, autonomous checkout is permitted only under Walmart's conditions. Consumers must authorize the agent through their Walmart account and establish permissions and spending limits, which Walmart enforces. To use the Gateway, the consumer must connect their agent to their Walmart account by authenticating directly with Walmart. So Walmart continues to accumulate user-level purchase data for use in a) the product recommendations returned by Sparky and b) Walmart Connect.
Third, third-party agents are restricted from using Walmart’s data for any purpose that could possibly intermediate Walmart. From its terms:
An Agent or its Deployer must not: (a) conduct systematic or bulk scraping, collection, storage, or other use of Walmart Proprietary Data; (b) retain Walmart Proprietary Data beyond what the Permitted Purpose reasonably requires; (c) use Walmart Proprietary Data to generate content, target advertising, create synthetic data, train models, or produce commercial analytics, reporting, or data-licensing products; or (d) license, resell, syndicate, or redistribute Walmart Proprietary Data to any third party.
Walmaet’s agent gateway relegates external agents to distribution interfaces, and it preserves Walmart's first-party consumer relationship and advertising economics. This isn’t a strategic embrace of third-party agentic commerce.
I was going to publish about this last week but got sidetracked by $META earnings. Post going live tomorrow.
6
1
30
3,708
It was truly inspirational to watch @NStoronsky operate closely. He's one of the most ambitious founders I have met. During the time I was at @Revolut, he would be the first person in the office and the last to leave. Ungodly amount of energy. He would meet all the teams on a regular basis and provide clear feedback on every product.
We brought several learnings with us to @sardine, several of them were very counter intuitive at the time!
We brought several learnings with us to @sardine, several of them were very counter intuitive at the time!
Nik Storonsky is one of the most hardcore and competitive founders I’ve ever talked to. He’s the founder of Revolut, a $115 billion company, and wants to be the number one bank in 100 countries.
When someone asked Nik if he was coming for Jamie Dimon's lunch, he said “I’m coming ant to dominate, otherwise what’s the point?”
He removed every VC from his board, and when one investor pushed him too far, he started his own fund to compete with them.
Nik has a lot of ideas on talent, product design, marketing, constraints, and what makes founder-led companies special.
This is a fascinating conversation about competition, speed, and building a machine-like company.
0:00 Coming for Jamie Dimon's dinner
2:04 Building the #1 bank in 100 countries
6:12 Why founders beat professional CEOs
8:45 Revolut's biggest mistake
12:47 Why legacy banks can't innovate
17:20 Running a company like a machine
20:28 Funding winners and cutting losers like a trader
23:31 How AI is changing Revolut's team structure
25:23 50 direct reports and spreading innovation fast
29:59 Hiring self-guided missiles
33:31 Too much money makes founders weaker
37:38 What Europe's constraints taught Revolut
41:08 Why VCs are like football cheerleaders
49:28 Starting a fund to compete with VCs
54:28 From zero marketing to buying Europe's airports
1:00:21 Funding moonshots like a global tech company
1:03:58 Building an unstoppable machine
Includes paid partnerships.
5
4
38
4,684
Jack Dorsey and Block have released a new voice model that feels fully conversational. It is small enough to run on a phone. This is going to help with customer service for sure but also scams are going to proliferate even more.
Most bank fraud and scams start from a phone call
Because a bank's phone number can be easily spoofed as most telcos don't support STIR/SHAKEN to authenticate callers.
We have a product called Sardine Shield to help you identify scam calls. Currently the Android app is available in India but coming soon to other markets:
sardine.ai/shield
1. When you get a call from a phone number not in your address book, it risk rates that call based on caller reputation
2. Next week, we are releasing a version where high risk calls will be diverted to an AI model.
3. The AI model then has a conversation with the potential scammer (perhaps also using AI).
4. If we think it's a definite scam, the call is dropped else handed over to human to continue.
Our app has been downloaded 50k+ times, it's free and we want more consumers using it to protect themselves.
If you are in India, give it a spin.
If you want early beta for the AI call diversion, comment below.
Would love feedback as you try it!
Most bank fraud and scams start from a phone call
Because a bank's phone number can be easily spoofed as most telcos don't support STIR/SHAKEN to authenticate callers.
We have a product called Sardine Shield to help you identify scam calls. Currently the Android app is available in India but coming soon to other markets:
sardine.ai/shield
1. When you get a call from a phone number not in your address book, it risk rates that call based on caller reputation
2. Next week, we are releasing a version where high risk calls will be diverted to an AI model.
3. The AI model then has a conversation with the potential scammer (perhaps also using AI).
4. If we think it's a definite scam, the call is dropped else handed over to human to continue.
Our app has been downloaded 50k+ times, it's free and we want more consumers using it to protect themselves.
If you are in India, give it a spin.
If you want early beta for the AI call diversion, comment below.
Would love feedback as you try it!
We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.
block.xyz/dessa
2
1
17
2,111
Great take on personal agents. Also can attest to the fact that when agents spin up a Virtual Machine to run a task, then while making a payment they often get blocked.
Because the IP address for most VMs is that of a data center - since the VMs are running there
We have a solution we call Know Your Human (KYH) that solves for this
Will be sharing this next week. Comment if you want to collaborate with us:
- if you are at a merchant who wants to accept agent traffic
- if you want to be the first agent company with the highest auth rate
- if you are building a wallet for agent companies and want to provide the best auth rates
Because the IP address for most VMs is that of a data center - since the VMs are running there
We have a solution we call Know Your Human (KYH) that solves for this
Will be sharing this next week. Comment if you want to collaborate with us:
- if you are at a merchant who wants to accept agent traffic
- if you want to be the first agent company with the highest auth rate
- if you are building a wallet for agent companies and want to provide the best auth rates
My unfiltered thoughts on personal agents.
Grokbot has the best UI and it’s arguably the best overall. In a lot of ways they’re ahead. Their desktop app feels the best, and their multi-bot coordination is great: you can set a team lead and it passes tasks to the other bots. Grokbot also has the best “webhook” feature, which means you can easily turn outside events into triggers that start your agents. That speeds up background workflows like watching your email, Slack and more. It also has Cursor integration, so for bigger tasks Grokbot can spin up a Cursor cloud session with any model (until OpenAI cuts SpaceX off from their models, which I think will happen soon). Unfortunately, running frontier models through Cursor costs more than Claude or Codex. Still, they’re the only ones embracing a team of agents. It’s a really cool idea, but the more bots you have, the more time you spend managing and maintaining them. I’m starting to think one bot with different threads/channels is the right answer, which I’ll get to with Claude Duet.
dots by OpenAI is rough around the edges right now, but it’s probably the most powerful if you use Codex, which I do. dots can spin up Codex sessions in the cloud or locally on your computer. So if you have coding or video editing workflows that need your local files, this is very useful. Codex computer use is also the best in the world: your dot can spin up a local Codex session that fully controls your computer, as long as it’s on. Their real-time voice is incredible. If you have your email and work tools connected, try this: go on a walk with headphones in, ask the agent to summarize your emails, have it draft replies and read them back to you, then send them off. If I remember right, this is close to a workflow you (Tim) used to have with your VAs before AI. It’s my favorite use case for personal agents. The one downside is that you have to call your agent, and it’s an awful experience. It literally rings, and it’s annoying.
Muse by Meta is a toy. I’ve talked to a bunch of finance people I play poker with who all tried it and set up automations, and they all said it doesn’t reliably do what they want. It’s fast, and its user numbers will probably keep growing. But it’s basically free: you get a TON of usage without paying a dime. That’s NOT a trade I’m willing to make. An agent doing high-leverage tasks should be running at the Opus 5.5 or GPT 6.1 Sol level.
Claude Duet - is in early access, so unfortunately not everyone can use it yet. First things first, it runs on Opus 5.5, the best model in the world. I’ve only had access for 72 hours, but Duet feels different from the others. It’s not a personified agent like Grokbot, dot or Muse, which gives it a more professional vibe that I like. When you ask your Duet agent to do something, it spins up a task, which is a new chat thread. All your threads show up in a Tasks tab, so it’s easy to see everything Duet is working on or has finished. And instead of creating more agents (like Grokbot), you create channels, just like Slack (I can tell @anthropic is Slack-pilled). I have a content channel, a finances channel, etc., which keeps different kinds of tasks separate. The only other app I’ve seen do this is Buzz by @jack, and it’s a good experience. Their strong point is documents (artifacts). If you want decks, high-quality docs or design mockups, Opus just wins. Period. It’s not close right now. Their weak point is real-time voice. It’s the worst of the bunch and borderline unusable, though I think they only added it a few days ago.
Instinct lives in iMessage, has no app, and doesn’t do anything notable besides show you product suggestions you didn’t ask for. (Not ads, according to their team.)
My hot take:
The VM is overrated. SpaceX, OpenAI and Meta all talk about the VM your agent can use and feature it in their marketing. It’s not as useful as you’d think, because more and more websites block bots that visit their site on the cloud VM. All of these companies promise their agents can make payments, but most payments either take forever or get blocked altogether. SpaceX has a cool workaround with Grokbot that routes traffic through your computer so it uses your IP address. But I’d much rather the agent just USE my computer for the task, like dot does by handing it off to Codex. Also, i don't think normal people care about the virtual computer. If anything they get offended they have to "stream a computer".
The problem with personal agents: by the time you read this, pick one, and test it for 7 days, every one of them will have released 10 new features and the whole landscape will have changed. It’s nearly impossible to keep up.
5
1
24
5,031
🚨 Excited to share Sardine's latest product suite on AI Token fraud: sardine.ai/token
AI Token is the new gold bar that fraudsters and nation-states steal and launder to distill their own models. Here's how Sardine helps:
1. Bank-grade "Know Your Business" (KYB) and "Know Your Consumer (KYC), so you are not onboarding shell companies set up to distill frontier models
Sardine offers KYC and KYB globally. The same bank-grade technology that protects financial institutions globally from illicit activities is what you get.
2. Signup fraud: We want you to offer free trials and Token credits without losing all those credits to fraud.
The same fraud prevention technology that helps the top fintechs, top crypto companies, top prediction markets and top social media companies, is what you get to protect your credits.
3. Account takeovers that lead to Token theft: We protect API keys to your models or to your 3rd party frontier labs with the same security posture as we protect crypto assets at top crypto exchanges.
We have a two-tiered approach. Detect if a new IP was used to query an API key. Challenge such requests by having the user do a login (where we run our device fingerprinting). We have the fastest growing fraud device consortium (6B+ devices profiled globally).
Fraud and money laundering rings have a nexus - the same rings that steal crypto, gift cards, cars, fintech wallets are the ones going after AI Tokens.
4. Card fraud: Fraudsters use stolen cards to buy AI Tokens, that they can then quickly resell.
The same technology that protects the top retailers from Card fraud is what you get!
Fraudsters method don't change, only what they steal changes:
2005-10: Smart TVs and physical goods
2010-15: Gift cards
2015-25: Crypto
2025: AI Tokens
AI Token fraud has proliferated across the AI supply chain and impacting AI inference companies, neoclouds, AI gateways. In the last weeks alone, OpenCode, Railway, LiteLLM and many others have been impacted.
We are here to protect you and to help you win back trust from your customers. Reach out. DMs open!
AI Token is the new gold bar that fraudsters and nation-states steal and launder to distill their own models. Here's how Sardine helps:
1. Bank-grade "Know Your Business" (KYB) and "Know Your Consumer (KYC), so you are not onboarding shell companies set up to distill frontier models
Sardine offers KYC and KYB globally. The same bank-grade technology that protects financial institutions globally from illicit activities is what you get.
2. Signup fraud: We want you to offer free trials and Token credits without losing all those credits to fraud.
The same fraud prevention technology that helps the top fintechs, top crypto companies, top prediction markets and top social media companies, is what you get to protect your credits.
3. Account takeovers that lead to Token theft: We protect API keys to your models or to your 3rd party frontier labs with the same security posture as we protect crypto assets at top crypto exchanges.
We have a two-tiered approach. Detect if a new IP was used to query an API key. Challenge such requests by having the user do a login (where we run our device fingerprinting). We have the fastest growing fraud device consortium (6B+ devices profiled globally).
Fraud and money laundering rings have a nexus - the same rings that steal crypto, gift cards, cars, fintech wallets are the ones going after AI Tokens.
4. Card fraud: Fraudsters use stolen cards to buy AI Tokens, that they can then quickly resell.
The same technology that protects the top retailers from Card fraud is what you get!
Fraudsters method don't change, only what they steal changes:
2005-10: Smart TVs and physical goods
2010-15: Gift cards
2015-25: Crypto
2025: AI Tokens
AI Token fraud has proliferated across the AI supply chain and impacting AI inference companies, neoclouds, AI gateways. In the last weeks alone, OpenCode, Railway, LiteLLM and many others have been impacted.
We are here to protect you and to help you win back trust from your customers. Reach out. DMs open!
1
7
40
3,164
While the past decade was about crypto token fraud and security. The next one will be about AI token fraud.
At its core, API keys are like credit cards or a crypto asset.
The worlds of crypto and credit cards are way more advanced when it comes to securing them. In crypto, we e (MPC) wallets, Shamir secret sharing, etc. In credit cards, we have card tokenization so you can hide the real card number. In the world of API keys, we need something equivalent.
Also the same fraud systems we use to secure and protect crypto tokens and credit cards, can be applied to protect AI tokens.
Here’s a reference design for how neoclouds and AI gateways should think about API key security:
1) First, build a relationship between an API key and IP-addresses used to access it. For example, most developers who use coding harnesses, are likely to only access the frontier model API via a couple of trusted IP addresses - their residence and office.
Whenever a new IP is found to be accessing an API key, from an abundance of caution, you can revoke that API key immediately.
2) Next, you can then force the user to log back in to generate new API keys. When they log back in, that’s when you run a device fingerprinting check on their browser, and establish whether they are coming from the same trusted device fingerprint you associated with them before.
This helps you establish if in Step (1) prior, whether the developer was traveling or had a new laptop vs. whether their API key got compromised
Open questions:
1) I am not an expert in the world of DPoP, Biscuits and mTLS based API keys. Would love to learn from others. I can pattern match and see there's a huge potential to do something in API key security!
2) To everyone who works at neoclouds and AI companies - keep up the great work fighting the bad guys. We @sardine are here to help protect you from fraud. DMs open.
At its core, API keys are like credit cards or a crypto asset.
The worlds of crypto and credit cards are way more advanced when it comes to securing them. In crypto, we e (MPC) wallets, Shamir secret sharing, etc. In credit cards, we have card tokenization so you can hide the real card number. In the world of API keys, we need something equivalent.
Also the same fraud systems we use to secure and protect crypto tokens and credit cards, can be applied to protect AI tokens.
Here’s a reference design for how neoclouds and AI gateways should think about API key security:
1) First, build a relationship between an API key and IP-addresses used to access it. For example, most developers who use coding harnesses, are likely to only access the frontier model API via a couple of trusted IP addresses - their residence and office.
Whenever a new IP is found to be accessing an API key, from an abundance of caution, you can revoke that API key immediately.
2) Next, you can then force the user to log back in to generate new API keys. When they log back in, that’s when you run a device fingerprinting check on their browser, and establish whether they are coming from the same trusted device fingerprint you associated with them before.
This helps you establish if in Step (1) prior, whether the developer was traveling or had a new laptop vs. whether their API key got compromised
Open questions:
1) I am not an expert in the world of DPoP, Biscuits and mTLS based API keys. Would love to learn from others. I can pattern match and see there's a huge potential to do something in API key security!
2) To everyone who works at neoclouds and AI companies - keep up the great work fighting the bad guys. We @sardine are here to help protect you from fraud. DMs open.
1
2
12
556
Crazy day in the world of security today. From 2015-25, security loopholes were exploited to go after crypto tokens. Now it's AI tokens.
The same attackers that went after @opencode (popular open source coding harness)
are now going after Railway (host for AI apps) and LiteLLM (popular open source router)
And all of them are after the frontier labs API keys, so they can distill their models
The same attackers that went after @opencode (popular open source coding harness)
are now going after Railway (host for AI apps) and LiteLLM (popular open source router)
And all of them are after the frontier labs API keys, so they can distill their models
3
0
17
1,302
Very similar conversations we have been having with everyone around Agentic commerce. Know Your Human (KYH) is what is actually needed by all participants around Agentic commerce - to really identify the human behind the agent. More to come soon in a white paper we will publish next week
M2020, Recap Part 3. Red hot bank issue is Meta Muse in "Agentic Finance". Per previous tweet, consumers not only giving over their banking credentials, but also their OTP. Expect move to structured APIs, explicit consumer permissions that banks hold, data agreements between agentic platforms and banks (with rules and pricing).
Most agree that KYA makes no sense. Banks don't need to know the agent, they need to know what the consumer authorized, and how the consumer was authenticated (by the agent). They also need ensure the agent operator will abide by access terms and rules (ex never asking for OTP).
9
1
33
3,776
This is the future of security and anti fraud. We built @sardine with those exact principles in mind:
Stop fraud at machine speed without waking up humans
Our device fingerprinting product detects faudulent accesses and blocks them automatically
We have a Graph Agent that traverses the full graph of users connected to each other because of shared identifiers (same device id, residential IP, email etc)
So in one click you can take out the entire fraud ring!
And a Data Analyst Agent that analyzes access patterns and suggests new fraud rules
These Agents run as a scheduled Task 24/7/365 protecting you.
While your fraud & security team sleeps
Stop fraud at machine speed without waking up humans
Our device fingerprinting product detects faudulent accesses and blocks them automatically
We have a Graph Agent that traverses the full graph of users connected to each other because of shared identifiers (same device id, residential IP, email etc)
So in one click you can take out the entire fraud ring!
And a Data Analyst Agent that analyzes access patterns and suggests new fraud rules
These Agents run as a scheduled Task 24/7/365 protecting you.
While your fraud & security team sleeps
they've been non stop trying to attack our systems for the past 48 hours
a side effect is everything is getting hardened but damn we basically need agents going 24/7
because that's what they have
1
0
14
866
Soups Ranjan retweeted
We had a security breach last night related to a large scale distillation attack that impacted 1.4% of OpenCode Console users
We have notified them all this morning via email. If you did not get an email you were not impacted, it was a subset of users who had external providers linked
The incident is ongoing so we cannot share more details at this point but we are working with Anthropic, OpenAI, and Google to remediate
We have notified them all this morning via email. If you did not get an email you were not impacted, it was a subset of users who had external providers linked
The incident is ongoing so we cannot share more details at this point but we are working with Anthropic, OpenAI, and Google to remediate
738
1,475
11,658
2,964,262
.@money2020 is upon us. A lot has changed in the world of fraud and payments in the last 7 years we've been going.
A sample of hot topics we @sardine discussed in the past years and what we are focused on this year:
2026:
* Job applicant fraud
* Agentic commerce fraud
* AI Token fraud
2024:
* Faster payment fraud (RTP/Fednow, Zelle)
2020:
* Crypto fraud
* Fintech account funding fraud
A sample of hot topics we @sardine discussed in the past years and what we are focused on this year:
2026:
* Job applicant fraud
* Agentic commerce fraud
* AI Token fraud
2024:
* Faster payment fraud (RTP/Fednow, Zelle)
2020:
* Crypto fraud
* Fintech account funding fraud
3
3
22
1,361
If the device/browser is indeed something you’ve seen before, then you know it’s a false positive - perhaps because the user is traveling or got a new machine. Then you let them create a new API key.
If the device/browser was never seen before, then you know something’s off and perhaps that user’s API key got stolen – you could challenge them via many other step-ups thereafter ranging from $0 auth via a card, selfie verification, etc.
If the device/browser was never seen before, then you know something’s off and perhaps that user’s API key got stolen – you could challenge them via many other step-ups thereafter ranging from $0 auth via a card, selfie verification, etc.
1
0
2
101
If you work at an AI company - neocloud, AI gateway, coding harness - please reach out to us @sardine
AI Token fraud could take many different forms:
- Promo abuse
- Stolen card fraud
- API key theft
We can help you with all of these.
DMs open
AI Token fraud could take many different forms:
- Promo abuse
- Stolen card fraud
- API key theft
We can help you with all of these.
DMs open
0
0
4
121
2. Monitor access patterns for the various API keys and build an understanding of IP-addresses and device identifiers that should be considered trustworthy when associated with an API key
1
0
1
97
3. Most AI companies (e.g. coding harnesses, AI gateways) don’t offer a UI - so it’s difficult to fingerprint their device or browser. However, there’s a two-step process you can follow:
First, if you monitor an API key is being accessed from a brand new IP-address never seen , you could revoke the API key; and
Second, send an email to the user to log back into your product, at which point you fingerprint their device/browser.
First, if you monitor an API key is being accessed from a brand new IP-address never seen , you could revoke the API key; and
Second, send an email to the user to log back into your product, at which point you fingerprint their device/browser.
1
0
1
84
AI Token fraud is one of the fastest growing new fraud trends we are hearing about from neoclouds, AI Labs and AI routers.
Most AI companies provide some version of LLM routing – where you store API keys to various LLM models (Anthropic, OpenAI, etc) on behalf of your customers.
AI tokens are a great store of value, and easily fungible and easy to resell. Hence, they make for a very attractive target to steal - right up there alongside crypto, Stablecoins and gift cards.
Most AI companies provide some version of LLM routing – where you store API keys to various LLM models (Anthropic, OpenAI, etc) on behalf of your customers.
AI tokens are a great store of value, and easily fungible and easy to resell. Hence, they make for a very attractive target to steal - right up there alongside crypto, Stablecoins and gift cards.
1
0
1
208
Trust is very fragile: takes years to earn and one mistake, in one second and you lose it.
An AI company that takes security and privacy very seriously, wins in the long run. Here are some thoughts on what AI companies could do to bolster their security:
1. Never store API keys (for eg to access models from frontier labs) in cleartext. Use a key management service to decrypt them in memory before using them
An AI company that takes security and privacy very seriously, wins in the long run. Here are some thoughts on what AI companies could do to bolster their security:
1. Never store API keys (for eg to access models from frontier labs) in cleartext. Use a key management service to decrypt them in memory before using them
1
0
1
102