Snippings

Simon Taylor

the merchants supporting agents are showing high ROI, the average Stripe Link instinct user spends $1300 per month

From: Home / X.

xxx

xxx

Hackers behind the record-breaking Binance heist spent weeks masquerading as recruiters and fake executives, eventually convincing an engineer to download malicious software onto a work computer as part of a larger-scale ruse, the company said Tuesday.

From: Fake job interviews helped hackers pull off Binance’s record crypto heist.

xxx

xxx

Accenture co-authored a report about artificial intelligence that was littered with false citations, incorrectly attributed research and case studies that appear not to exist, prompting accusations that the consultancy published information marred by AI hallucinations.

From: Accenture accused of publishing AI paper littered with AI hallucinations.

xxx

Banks can stick their heads in the sand and try to lock the stablecoin door, but that is not a sustainable long-term strategy. On the other hand they can look at a kind of Walmart approach to embrace and extend. Just as Walmart are XXX, they could build an AI-first brand within the bank or create an AI-first subsidiary. Adam Riccoboni makes an interesting alternative proposal though, suggesting that banks meet an industry-wide threat with an industry-wide response. He says that American banks should get together, pool resources, deploy an open-weight model such as DeepSeek-R1 and develop it into the world’s leading banking AI, owned by the banks and sold to banks worldwide. He draws on illustrative precedenta to make the point. In 1973, 239 banks from 15 countries created SWIFT, which now connects more than 11,500 institutions, and seven major American banks own Early Warning Services, the company behind Zelle. As he puts it, competing on the front end while sharing the plumbing is what banks have always done.

(Banks cannot pool customers’ private data though. Federated learning lets a model learn from information held in different banks without the raw data ever leaving each bank’s own system. Regulators will want to see how that learning is permissioned and locked down, but it is proven technology, and there are governance structures such as data trusts that can satisfy both privacy, security and operational requirements.)

xxx

More than half of American consumers, 53%, asked an AI for financial advice in the three months to March 2026. Only 47% of bank customers could recall getting any advice from their bank at all.

From: AI is coming for banks’ business, but the industry can fight back | American Banker.

xxx

xxx

Where an AI system is used within a financial institution, particularly where it supports an important or critical function, it may fall under both AI Act requirements and institution’s DORA risk-management framework. DORA addresses digital operational resilience and ICT risk while the AI Act focuses on trustworthy risk-based AI. Consequently, banks should treat critical AI Systems as ICT assets in the technology inventory and subject them to resilience testing, incident management and business continuity planning, including regulatory reporting where applicable. Supplier due diligence must also cover model governance, data provenance and security, and AI leads should work with enterprise risk, security and procurement to enforce model lifecycle controls through existing critical technology processes.

From: How the EU AI Act and DORA are changing AI governance in banking – GlobalData.

xxx

xxx

Then there’s Natural, a payment startup backed by venture firm Forerunner that has raised over $40 million in total. Natural is bringing banking products such as payments and credit lines to AI agents. Natural isn’t a bank but it’s offering a way for digital wallets used by agents to hold money in an FDIC-insured account. Natural uses Column Bank to hold the money, and each wallet is a sub-ledger into customers’ individual accounts.

Natural in turn offers its digital wallet to AI companies to use in agents, enabling the agents to send and request money. For example, Natural works with Conduit, a startup that offers AI agents used by short-term rentals and hotels to help manage their businesses. With the digital wallets, Conduit’s agents can accept payments, pay vendors and contractors, and store money.

From: How Smaller Banks Are Giving Agents Access to Accounts — The Information.

xxx

When it comes to letting your AI agent—such as Muse or Instinct—manage your personal finances or go shopping, your choice of bank will make a big difference. Some of the biggest banks in the U.S., like JPMorgan, are restricting AI agents’ access to consumers’ personal bank accounts, while some smaller banks are going in the opposite direction, giving AI assistants more access to consumer banking which they see as a potential area of growth. That includes Lead Bank, a Kansas City, Missouri-based bank that works with fintechs like Stripe, a major payment provider for AI agent apps. Lead Bank allows its fintech partners to issue bank account numbers for customers, reserved for AI agents. That in turn enables the agents to perform banking tasks like moving money through ACH and wire transfers while allowing the customers to limit what the agents can do in their accounts.

xxx

Brazil’s Mercado Bitcoin halts withdrawals after $40 million drained across 10 chains
The Brazilian exchange paused deposits and withdrawals after wallets on at least 10 chains were emptied, days after the $6.3 billion Binance hack.

From: Mercado bitcoin hit by $40 million hack across 10 blockchains.

xxx

Earlier this year there was a massive theft of cryptocurrency from the supposedly secure offline hardware crypto wallet Coldcard, made by Coinkite. The cirminals made off with around $130 million. This theft attracted attention because the point of using a product like Coldcard is that it’s supposed to be, at least in theory, one of the safer ways to store their cryptocurrency.

(The hackers figured out that there was a flaw in how Coldcard wallets generated users’ seed phrases, which made them predictable. Once they figured out the flaw, they simply needed to brute-force to obtain the victims’ seed phrases in reasonable time.)

An even more ingenious attack hit the Ledger hardware wallet. Hackers physically implanted hardware into Ledger wallets sold through Southeast Asian reseller CryptoBilis, and then used this trojan horse chip to get hold of the wallet contents, draining roughly $86 million from victims.