This Past Month In Bitcoin

AI Is Moving Faster Than Anyone Is Prepared For

The single most important data point this month came from METR, a nonprofit focused on evaluating AI capabilities. Their benchmark measures how long it would take a skilled human expert to complete a task that an AI agent can now do. Claude Opus 4.6 now has a 50% chance of completing a software task in seconds that would take a human 14.5 hours (METR Time Horizons, METR on X). The confidence interval is wide (6 to 98 hours) because METR's task suite is nearly saturated, but the trend line is clear: AI capability is doubling roughly every four months, with an R-squared of 0.93. Jack Mallers framed this starkly on his show: in four months, it will take seconds to do what takes a human more than a day. In eight months, half a week. The rate of change is exponential and it is not slowing down.

The real world consequences are already landing. Block (formerly Square) laid off over 4,000 employees, nearly half its workforce, citing AI directly (CNBC, CNN, Fortune, TechCrunch). The stock price soared 24% on the announcement. As Mallers pointed out, in a hyper-financialised economy, the market rewards companies for replacing humans with machines. Job openings in finance, insurance, and information sectors have fallen to levels not seen since the aftermath of the 2008 financial crisis, according to Bureau of Labor Statistics data via FRED.

Meanwhile, the Federal Reserve's Waller stated that reports on AI are overstating the risk to jobs. Multiple podcast hosts compared this to Ben Bernanke calling subprime "contained" in 2007. The pattern of institutional denial followed by crisis is familiar territory.

For Bitcoin businesses specifically, AI adoption is splitting along two distinct lines. On one side you have builders like Eric Cason, whose company Vora is creating a self-hosted AI platform where a local "Guardian" AI lives on your hardware, never touches the open internet, and is separated from an "Explorer" AI by an intermediary layer designed to prevent prompt injection and data exfiltration. On the other side you have operators like Ben at Low Wood Bay Resort, who has deployed corporate ChatGPT and Microsoft Copilot accounts across his management team, is tracking AI referral traffic to his website, and is building a generative engine optimisation strategy to make his content machine-readable for AI crawlers. Both approaches are rational given their contexts. But only one has examined the privacy implications.

The Privacy Crisis Nobody Is Talking About

Rabbit Hole Recap 398 surfaced two pieces of privacy intelligence that should concern every Bitcoin holder and every business in the space.

AI de-anonymisation is here. A research paper titled "Large-Scale Online Deanonymization with LLMs" (arxiv.org, The Register, author's writeup) demonstrated that AI can identify anonymous users at scale simply through writing patterns, timing analysis, and behavioural fingerprints. The researchers showed their agent could re-identify Hacker News users and interview participants from anonymous profiles and conversations alone, matching what would take a dedicated human investigator hours.

Matt Odell and Marty Bent connected this directly to Bitcoin. Chain surveillance companies are already using LLMs to analyse on-chain data. The terrifying implication is that the Bitcoin blockchain is permanent. A privacy mistake you make today, or made five years ago, can be exploited by future AI capabilities that do not yet exist. What passes for adequate operational security today may be trivially broken in two or three years.

Anthropic admitted it can identify users through their prompts. In a blog post about industrial-scale distillation attacks from DeepSeek, Moonshot AI, and Minimax (Anthropic blog, CNBC, TechCrunch), Anthropic revealed it had identified specific Chinese researchers by name through their search history and prompt patterns. Even when accessed through a proxy, the content and phrasing of your prompts can de-anonymise you. This is not theoretical. This is a company telling you, publicly, that they can do this.

The DJI backdoor story added another dimension. A user employed Claude Code to connect to their DJI Romo robot vacuum and accidentally discovered master backdoor keys to roughly 7,000 devices across 24 countries, all equipped with cameras mapping the interiors of people's homes (Popular Science, The Verge original, Tom's Hardware, Malwarebytes, Fortune). The RHR hosts connected this to Ring cameras, Flock surveillance cameras proliferating in parking lots and streets, and the broader pattern of consumer devices creating comprehensive surveillance infrastructure. Mix this with LLMs and the result is a panopticon that would have been science fiction five years ago.

Russia announced a 2.27 billion ruble ($29 million) programme to build AI-powered internet censorship, using machine learning to instantly block mirror sites and identify the people creating them (Ukrinform, UNITED24 Media, Reporters Without Borders). This is the authoritarian use case in its purest form, and the technology powering it is the same technology available to every government on earth.

The Zoom deepfake scam targeting Bitcoiners was flagged as an active, ongoing threat on RHR. Attackers compromise legitimate journalist accounts (including the official New York Post account), set up Calendly meetings, then play deepfake video of the person you think you're talking to while the "audio doesn't work." While you troubleshoot, you install malware. Multiple listeners reported friends being hit in the last week.

Who Is Building What

Low Wood Bay Resort and English Lakes Hotels

Ben's interview provided a detailed look at how a fifth-generation family hospitality business is integrating Bitcoin. English Lakes Hotels operates multiple properties in the Lake District including Low Wood Bay Resort (the flagship), Lancaster House, and Wild Boar Estate. Ben took over as Managing Director in 2022.

The Bitcoin integration is genuine but faces real-world friction. They accept Bitcoin payments on their website and in-person, and crucially, they hold all Bitcoin received rather than converting to fiat. They are building toward a circular economy by sourcing coffee directly from a plantation in El Salvador, paying the farmer in Bitcoin with no intermediaries and no Swift fees. They educate staff using HRF videos and copies of The Little Bitcoin Book.

However, the ambitions to pay staff in Bitcoin and offer Bitcoin pension top-ups have been blocked by UK legal advice around inducing investment. Capital is being deployed into physical expansion (spa, lakeside sauna, new experiences) rather than Bitcoin treasury because the returns from business investment currently exceed holding Bitcoin on the balance sheet.

On the AI front, Ben has made it a standing board agenda item, split across systems (agent automation, menu checking), sales (generative engine optimisation), and personnel (AI 101 training programme, corporate ChatGPT and Copilot accounts). He is tracking referrals from AI tools to their website, which are doubling monthly from a small base. This is forward-thinking from a business perspective but entirely unexamined from a privacy perspective.

Vora (Eric Cason)

Eric Cason's Vora represents the opposite end of the spectrum. It is a hardware-based, self-hosted AI platform designed from the ground up around privacy and data sovereignty (vora.io, book: cryptosovereignty.org). The architecture separates a local "Guardian" AI (which never touches the internet and stores all personal data) from an "Explorer" AI (which goes online to gather information), with an intermediary layer filtering data between them to prevent prompt injection and data exfiltration.

The longer-term vision involves a CT Network where Vora devices communicate with each other using homomorphic encryption, with backup infrastructure running on Lightning for payments and Nostr for identity. If your device is stolen, you invoke a recovery protocol, pay the network via Lightning, and download your encrypted data.

Vora is still in alpha and moving slower than the team would like, but their thesis is being validated weekly by stories like the DJI backdoor. Eric frames data ownership as equivalent to protecting your Bitcoin private key, arguing that Bitcoiners are the only community with the ethos to understand why this matters.

Strike

Jack Mallers announced two major developments. First, Strike obtained its New York BitLicense after nearly four years of work, opening the New York market. Second, Strike launched a perpetual Bitcoin-backed line of credit with bill pay and direct deposit integration (strike.me).

The "Smart Lock" strategy Mallers described is designed to let individuals live entirely on a Bitcoin standard. The concept: instead of spending your income on expenses and investing what is left over, you invest all income into Bitcoin and borrow against it at 13% APR to fund your life. As long as Bitcoin's compound annual growth rate exceeds 13%, your collateral grows faster than your debt. The tool keeps loan-to-value ratios below 15% to survive an 80% crash without liquidation.

This is infrastructure that Bitcoin businesses could also use, running their operations on a Bitcoin standard with line of credit financing for expenses and bill pay. Mallers explicitly noted the product works for businesses as well as individuals.

Bitcoin Mining Economics

Bob Burnett provided detailed intelligence on the mining landscape. The average global cost of production sits in the mid-to-high $80,000s per Bitcoin. The most efficient operators produce at $50,000 to $55,000. The least efficient, primarily large public mining companies, are in the low six figures. Global hash rate has been flat for six months after years of meteoric growth, as miners pivot energy infrastructure to AI workloads. Hash rate data can be verified at mempool.space or Clark Moody's dashboard.

Burnett argues this is healthy for Bitcoin in the long run. If only the largest, least efficient operators survived through sheer capital access, it would be centralising and fragile. He also flagged that public mining companies have been misvalued as tech and growth companies when they should be evaluated as commodity producers, measured by efficiency and productivity rather than growth metrics. Mining cost data can be cross-referenced with quarterly reports from public miners like MARA, RIOT, and CLSK.

Numopay / eCash Point-of-Sale

The Cashu development team launched Numopay, a free, open-source tap-to-pay Bitcoin point-of-sale app for Android (numopay.org, Bitcoin Magazine). It works with any NFC-enabled Android device, supports both Lightning and Cashu eCash payments, charges zero platform fees, and works offline. A BTCPay Server integration is in development. This is the kind of infrastructure that could enable businesses like Low Wood Bay to accept Bitcoin payments with a dramatically better UX than QR codes. Compatible Cashu wallets include Macadamia, eNuts, and others.

The Macro Backdrop: Fragility Everywhere

Multiple conversations this month converged on the same conclusion from different angles: the system is more fragile than most people understand, and the window for building alternatives is narrowing.

Jack Mallers walked through the fiscal doom loop in detail, drawing on data from Michael Howell's Cross Border Capital. True interest expense on US debt is approaching 100% of tax receipts. Credit card delinquencies are at 12.7%. Auto loan delinquencies are above 5%. Student loan delinquencies hit an all-time high of 16.3% (sources: Yahoo Finance, Bloomberg, Redfin, BLS data via FRED). Jamie Dimon publicly stated he is seeing parallels to the era before the 2008 financial crisis. There are 44% more home sellers than buyers.

Larry Lepard connected the macro picture to Bitcoin valuation. Using both the Tipmeyer multiple and the power law model (Giovanni and Kruger), Bitcoin at its current price has only been this cheap 6% to 10% of its history. The power law model suggests fair value around $120,000 to $130,000, with the historical floor at half a standard deviation below that ($50,000 to $55,000, which notably aligns with Burnett's most efficient mining cost). Lepard also noted the Fed has already begun expanding its balance sheet, with $525 billion added in three months and an annualised liquidity growth rate of 20% over the past six months.

ISM Manufacturing PMI came in at 52.4 (verify at ismworld.org), indicating expansion for two consecutive months. Historically, this has been a strong signal for the beginning of a Bitcoin bull market.

Eric Cason put it most bluntly: in five years, we either have freedom or we do not. Those are the options on the table. The technology for total surveillance exists. The technology for resistance also exists. Which one wins depends on what gets built and adopted in the next few years.

What to Watch Next Month

The AI de-anonymisation threat to Bitcoin businesses. This is the story that nobody in the mainstream Bitcoin media is covering with the depth it deserves. Every Bitcoin business feeding data into third-party AI platforms is creating behavioural fingerprints that can be used against them. We will be tracking which businesses are examining this risk and which are ignoring it.

Strike's line of credit rollout. If this product works as described, it is a fundamental shift in how Bitcoin holders and businesses interact with the fiat system. Watch for adoption numbers, which states get access, and whether businesses begin using it for operational financing.

Mining hash rate and the AI pivot. Six months of flat hash rate is significant. If miners continue redirecting energy to AI, it changes the security assumptions of the network and the economics of remaining miners. Watch for difficulty adjustment trends and public miner quarterly reports.

Numopay and eCash point-of-sale adoption. The launch of zero-fee, offline-capable, tap-to-pay Bitcoin payments for merchants could accelerate physical world adoption. Watch for the BTCPay Server integration and merchant adoption numbers.

Open source AI progress. The gap between convenient but surveilled AI (Claude, ChatGPT) and private but difficult self-hosted alternatives (OpenClaw, Vora) is the defining tension in freedom technology right now. Any narrowing of that gap matters enormously.