Gavin's Bindery

Today in "we're not ready" (no, really!): Unvetted AI agents are accessing U.S. bank accounts Major U.S. banks are detecting unauthorized third-party AI agents operating inside customer accounts https://www.axios.com/2026/10/22/ai-agents-banks-unauthorized-accounts-access

today in we're not ready: "sovereignty washing" is a phrase I'm now seeing a lot more. ... a marketing practice where cloud, software, or AI vendors falsely or misleadingly market their services as "sovereign," "locally controlled," or compliant with regional data laws while retaining foreign ownership, operational control, or legal exposure. Example: https://nextcloud.com/blog/sovereignty-washing-microsoft-cloud/ (I'm not commenting on Microsoft)

Today in we're not ready: Genetically Modified Military Dog Startup raises $200M to create “elite working dogs” equipped with "advanced body armor" https://www.wired.com/story/cool-cool-cool-palmer-luckey-made-a-genetically-modified-military-dog-startup

Today in "we're not ready" and O.M.G. The eight affected private keys were: 4, 56, 66, 115, 155, 156, 230, and 237 with some speculation that whoever created all the private keys misunderstood “create a 256-bit key” to mean “pick a value between 1 and 256" https://tech-insider.org/emudhra-ca-private-keys-compromised-number-4-2026/ (cf: https://xkcd.com/221)

REDSHIFT: Slow Down and Reconnect with Time, a new 20-minute immersive audio visual experience in collaboration with Tom Middleton @tommiddletonmusic https://binarydust.org/2026/10/17/redshift/

EVENT! Redshift: Tom Middleton & Gavin Starks present a 20-minute acoustic cosmology for our temporal restoration Details & tickets: https://binarydust.org/2026/10/17/redshift #ADE2026 #EarthSonic #amsterdamdanceevent #electronicmusic #soniverse #europeanspaceagency #ESA

A practical guidance for using AI transcription safely https://www.careful.industries/ai-voice-transcription

Here's a diagram I drew many, many years ago (inspired by many others). When we talk about "minimum viable product" and "agile development" we don't just mean 'iterating quickly', we encourage putting user needs at the heart of everything we do. What examples do you have?

Today in "we're not ready": JPMorgan CEO blocks outside AI agents from customer accounts "if customers choose to use outside agents, they should understand the liability transfer." https://www.invela.com/blog/jamie-dimon-its-about-who-bears-the-risk

Today in "we're not ready": One of the most advanced and wealthiest ai companies on Earth can't add up own revenues without a 50% error bar. It's $20B ($20,000,000,000) short. https://giftarticle.ft.com/giftarticle/actions/redeem/d77c36c8-336b-43c9-acef-f61b7a5507d9

Today in Hegel mirrors*: The 'data gap' is leadership. https://www.linkedin.com/pulse/billion-dollar-botch-how-government-bureaucracy-strangled-kam-rafiei-3navc/ * Hegel, 1837: "what experience and history teach is this, - that peoples and governments never have learned anything from history, or acted on principles deduced from it."

Today in 'actually, this bit was done already' or 'would you like a Hegel mirror'? Another week, another moment where I feel like I'm watching this behaviour as (yet) another group try to reinvent the internet and the web. [image shows a man whittling a wheel out of wood by hand in his shed while stating that he doesn't want to reinvent the wheel. Next to him is an entirely automated production line of wheels, mass produced, in better quality, for lower cost as part of an advanced industrialised society. Image is AI_generated; the ego of the man, however, is entire of his own making.]

Today in "billionaire can only think in billions": Bill Gates warns AI could cause ‘a billion deaths’ https://giftarticle.ft.com/giftarticle/actions/redeem/f4913aa5-6739-4cca-8dd9-a42c5a3727ba

AI systems inherit the governance failures of their training data. If the data is unverifiable, the model is unverifiable. If the licensing is unclear, every output carries legal risk. If there is no provenance, there is no audit trail. If there is no declared purpose, there is no basis for trust. https://ib1.org/ai

AI systems inherit the governance failures of their training data. If the data is unverifiable, the model is unverifiable. If the licensing is unclear, every output carries legal risk. If there is no provenance, there is no audit trail. If there is no declared purpose, there is no basis for trust.

The market cannot trust infrastructure it cannot inspect. Should trust frameworks be digital public infrastructure? https://ib1.org/trust-services/

Today in we're not ready: On a potential "US operation against a Chinese vessel"—"a chatbot the analyst used inaccurately identified the material the ship was carrying” [and] “almost started a war” a source said https://edition.cnn.com/2026/09/18/politics/us-military-ai-false-intelligence-china-ship

Today in we're not even at the start of all this yet are we? "If we build a superintelligence, we will first fall in love with it. Then we will worship it. Finally we will make sacrifices to it." https://futurism.com/future-society/cult-anthropic-insiders-worshipping-claude-god

Today in I really doubt we've thought this through. I think I'm more concerned about self-replicating tech-bros than I am about ai.

Today in 'we're not ready': How many questions you ask at once changes model accuracy dramatically. https://kevinmd.com/2026/09/why-prompt-engineering-in-medicine-beats-picking-a-model.html

Today in you're not ready: Your ai conversations may be being read by humans. I wonder what you will stop doing because of this? https://www.404media.co/inside-project-lily-the-humans-reading-your-chatgpt-chats/

discuss: "The UK has built the strongest innovation position it’s had since the late Victorian period" vs "is quietly handing its most valuable public intangible assets to (mainly) US tech companies for the equivalent of modern beads" https://articape.substack.com/p/act-v-the-country-as-a-p-and-l

Today in "we're not ready" "Anthropic Is Building a Predictive Surveillance System to Monitor Activists" https://prospect.org/2026/09/09/anthropic-artificial-intelligence-surveillance-system-monitor-activists/

IB1 response to DSIT’s Data regulation in the age of AI tl;dr - data governance is lagging AI capabilities; address with open standards, clear purposes, machine-readable rules, provenance and audit frameworks https://ib1.org/2026/09/10/response-to-dsits-data-regulation-in-the-age-of-ai/

Today in "this may resonate with some of you" Over the past NaN decades, I've heard people saying they don't want to re-invent the wheel, often while I can see them holding a piece of wood that they are whittling it into a wheel. Any tips from the frontline?