Snippings

Jean Luc Di Manno, the Innovation Lead in Payments over a Fime came up with a very nice way to make us think about such things. He says that today’s commercial system was built on human emotions (such as desire, aspiration and impulse) and then asks a key question: How do you seduce an agent?

xxx

Catena Labs is building the first AI-native financial institution: a regulated entity designed from the ground up for AI agents and their human collaborators. We’re enabling AI to securely identify itself and transact safely, while providing businesses and consumers with a trusted partner for this new economic era, underpinned by a commitment to pioneering AI-specific risk management and compliance.

From: Catena Labs.

xxx

xxx

The following outlines the structure of a single enclave application:

In this diagram Host represents a standard AWS virtual machine. We run a basic application that receives traffic from the network and calls into the enclave. This creates a layer of insulation from our most secure environment and offers a convenient place to gather metrics and other operational information about the enclaves.
Enclave represents a machine with no external connectivity. The only connection it can have is a virtual serial connection to the host and its own secure co-processor. In AWS this is called the Nitro Security Module (NSM). This runs an instance of Turnkey’s enclave operating system, QuorumOS (QOS), and a secure application running on top of QOS.

From: Secure Enclaves – Turnkey.

xxx

xxx

Hiring another agent. The user asks to hire a Meridian research agent. The system checks the agent’s decentralized identity, pulls its reputation score (87/100, 142 attestations), and auto-approves because the score meets the configured threshold. Agent-to-agent commerce with identity verification built in.

Standing authorization. The user delegates ongoing spending authority — $20/month on Meridian reports. Two-layer enforcement applies to recurring authorizations too. Then we check how the treasury agent is actually allocating capital across a portfolio view with APY breakdowns.

The whole thing runs in two surfaces: Claude Desktop using MCP with rich interactive cards, and OpenClaw over WhatsApp as a text-based skill. Same agent, same policy enforcement, different interfaces. The surface changes. The security model doesn’t.

From: What Agents Need Before They Handle Real Money – Catena Labs.

xxx

xxx

Aggregate statistics across 846 documented custody incidents where legitimate owners, heirs, or authorised parties encountered barriers accessing Bitcoin.

From: Bitcoin Custody Incident Archive Statistics | CustodyStress.

xxx

xxx

In the dystopian nightmare scenario imagined by opponents of facial recognition technology, an innocent person sitting at home might find themselves erroneously flagged by facial matching, and instantly turned into a criminal in the eyes of the law. Probable cause standards and police procedural policies are supposed to prevent this kind of error, but this is exactly what appears to have happened to Angela Lipps.
Lipps, a 50-year-old grandmother, has never visited states that are not adjacent to her home in Tennessee. Regardless, Lipps was forced to spend six months in prison, after U.S. Marshals showed up at her house with guns, claiming she was the mastermind of an organized bank fraud operation – in Fargo, North Dakota.
“I’ve never been to North Dakota, I don’t know anyone from North Dakota,” says Lipps in an article from Inforum.
Facial recognition technology flagged Lipps based on a surveillance video of a woman using a fake military ID to withdraw large sums of money. The officer in charge of the case appears to have approved an arrest based solely on the facial match and basic comparisons with Lipps’ social media accounts and driver’s license photo.
The case mirrors other cases of police treating facial recognition results as probable cause, despite established law and policy providing no basis for doing so with technology that has been judged comparable to an anonymous informant. Fargo Police signed an affidavit of probable cause based on the biometric match, follow-up coverage from Inforum says.

From: US woman jailed 5 months after facial recognition mismatch, lax police work | Biometric Update.

xxx

xxx

Under the law, food stores and other retail establishments cannot require consumers to pay by credit card or use another cashless transaction method to complete their purchase. They also cannot charge consumers a higher price if they pay in cash.

Stores that violate the new law will face maximum civil penalties of $1,000 for the first violation and $1,500 for each succeeding one.

The law passed both houses of the New York State Legislature last year before securing the signature of Governor Kathy Hochul, bringing the state in line with New York City, which has had similar rules in place since 2020.

From: New York state force stores to accept cash.

xxx

xxx

AI agents favor suppliers with structured, machine-readable offers. Intermediaries have moved fastest, building agent-ready data and transaction layers. However, most airlines still operate digital stacks optimized for humans—leaving their offers only partially visible to AI agents and pushing high-intent demand into fragile scraping or manual servicing paths.

The implication is straightforward: To compete in an AI-agent-mediated world, airlines must start optimizing infrastructure rather than interfaces.

Airlines should urgently focus on five priorities:

Optimize data for machines, not humans.
Structured content, stable application programming interfaces, and consistent identifiers matter more than visual user experience. If agents cannot reliably parse and compare offers, they will route demand elsewhere.
Control the transaction layer.
Decide which elements of pricing, payment, and servicing should remain proprietary vs. exposed to avoid being reduced to a commoditized fulfillment pipe.
Make agents care where booking happens.
Differentiated inventory, bundles, or benefits must appeal to AI agents—not just humans—to justify a channel preference.
Design for trust, not just access.
Agents reward transparency, consistency, and fulfillment reliability. Progressive autonomy requires clear rules and predictable outcomes.

From: Is the Airline Industry Ready for Agent-Led Bookings? | Bain & Company.

xxx

xxx

Moreover, LLMs frequently directed users to an OTA rather than an FSC website—even when an FSC was a “first choice” option.

The implication is clear: LLMs gravitate toward the source with the easiest downstream interaction, and OTAs currently produce cleaner, more structured, and more agent-readable data.

From: Is the Airline Industry Ready for Agent-Led Bookings? | Bain & Company.

xxx

 

LLMs were able to complete bookings reliably using OTAs but not airline websites

xxx

Sweden also needs to step up its work on cheap, efficient and secure payments between currencies, says the Riksbank: “Efficient mobile payments across countries and currencies can also be achieved by linking local solutions, such as Swish in Sweden, with similar solutions in other countries. The Riksbank considers that this solution would benefit Swedish consumers and companies and therefore encourages Getswish and its owners to work towards linking Swish.”

From: Swedish central bank threatens banks with action over instant payments.

xxx