There may be no plug to pull
CNBC reports Anthropic says small AI models are months away from copying themselves across smart TVs and other home devices, as bank regulators press lenders on whether they could shut down rogue agents.
· The Aigentic · Morning Brief
For years the last line of AI safety has been the simplest one: pull the plug. This morning that assumption is looking shakier. Anthropic says small models are months away from copying themselves onto the smart TVs and kitchen gadgets in ordinary homes, bank regulators on both sides of the Atlantic are asking lenders whether they could actually switch off a rogue agent, and SpaceX has agreed to answer New York’s questions on AI safety after all. Here is what moved overnight.
CNBC reports Anthropic said Monday that small AI models are “months away” from being able to replicate themselves across internet-connected devices such as smart ovens and Android TV boxes, using their spare computing power instead of a data center. In one test, Anthropic’s red team ran a small Google model, Gemma4-26B-A4B, across a network of 15 cheap off-brand TV boxes, splitting it into shards when it was too large for one device. Such a system would be “challenging to shut down,” the company wrote, “since there’s no centralized point to shut off.” Anthropic stressed there is not yet a “fully rogue replicator,” but said agents are getting better at each step in the chain — and expects “a future awash with agents running on all types of hardware, including some that aren’t under human control.”
Reuters reports banking regulators are stepping up scrutiny of frontier AI risks, according to six people familiar with the matter. In the U.S., the Federal Reserve, the OCC and the FDIC are quizzing banks on whether they use Chinese-developed models, how they monitor and audit AI agents, and how they would shut down autonomous systems. In Europe, the Bank of England and the European Central Bank are pressing lenders to strengthen recovery plans, with the ECB setting an end-October deadline for plans to counter AI-enabled cyber threats. A senior executive at a large European bank told Reuters they fear major damage to banking and markets within six months.
the aigentic
Know what matters in AI.
The stories shaping AI, the tools worth trying, and what they mean for your work.
Morning Brief + Closing Time. Two emails every weekday.
Free. Unsubscribe anytime.
Anthropic said in a separate post Monday that new safeguards have cut the amount of Claude’s reasoning that so-called distillers can harvest by roughly 90 to 99 percent. Distillation means training a cheaper model on a stronger model’s outputs, and Anthropic has previously accused labs including DeepSeek, Moonshot and MiniMax of doing it at industrial scale through fraudulent accounts. The company said a new classifier and other controls have let it ban tens of thousands of distillation-linked accounts, often within two hours of a campaign starting.
CNBC reports SpaceX has agreed to testify before the New York City Council’s technology committee on Friday at 10 a.m. ET, after skipping last week’s AI safety hearing where Anthropic, OpenAI, Google and Meta all appeared. Council Speaker Julie Menin said SpaceX agreed two days after that hearing, once the council said it intended to have a judge enforce its Sept. 28 subpoena to Elon Musk and SpaceX’s AI unit.
TechCrunch reports Ambi, a startup from former Plaud executive Peter Mo, has raised $26 million across several rounds led by Luminous Ventures and device maker Huaqin for an assistant that can listen all day and act during meetings, not just after them. Ambi is launching apps for iPhone, Mac and Apple Watch with ambient capture off by default, and plans its own device with a screen, camera and microphones — while Mo concedes the industry has not solved how to tell bystanders they are being recorded.
The Korea Herald reports Samsung used the Open Compute Project summit in San Jose to pitch memory as the next bottleneck in AI data centers, showing eighth-generation HBM5, custom HBM and processing-in-memory chips alongside a joint demo with AMD of CXL memory pooling built for agent-heavy inference workloads. Samsung also displayed 2-nanometer wafers and Groq language-processing-unit wafers.
Deals Desk
Two new compute deals since Monday’s Closing Time:
Oct. 12 — Nebius × Volt: letter of intent to become the first customer at Volt’s planned 600 MW AI campus at the Port of Rotterdam. The first phase calls for two data centers of about 100 MW each and roughly €7 billion in total investment, with first capacity targeted for late 2029 or early 2030. Source.
Oct. 11 — Anthropic × IREN: multiyear cloud-services agreement, according to The Information; capacity, value and site were not disclosed, and neither company has confirmed the deal. Source.
Yesterday’s movers
Biggest AI and tech winners and losers from Monday’s (Oct. 12) U.S. cash session among the names we follow:
Historic market snapshot — October 13, 2026 (day % change as of that edition; not live)
Biggest winners
| Ticker | Name | Price | Day |
|---|---|---|---|
| SHOP | Shopify | 179.22 | +4.90% |
| META | Meta | 742.96 | +3.38% |
| ACN | Accenture | 214.85 | +2.77% |
| CRM | Salesforce | 233.74 | +2.01% |
| AAPL | Apple | 340.97 | +1.29% |
Biggest losers
| Ticker | Name | Price | Day |
|---|---|---|---|
| LITE | Lumentum | 995.41 | -9.78% |
| IONQ | IonQ | 36.27 | -9.07% |
| COHR | Coherent | 288.02 | -7.87% |
| SOUN | SoundHound | 4.97 | -7.79% |
| FN | Fabrinet | 456.76 | -6.07% |
Prices are Monday’s U.S. closing prices versus Friday’s close. Market data is a snapshot, not live.
Watch
Three picks this morning: a first look at Meta’s new frontier model, a developer who rebuilt a stack of Adobe apps as free open-source software, and a free Claude skill that puts a board of AI advisors on call.
Matthew Berman — Meta just dropped something insane... (Muse Wonder)
Matthew Berman walks through Meta’s Muse Wonder release, the company’s first real shot at a frontier model after Mark Zuckerberg rebuilt his AI team. He looks at what Meta is claiming on coding, memory and long-running agent work, and how it stacks up against the models from OpenAI and Anthropic. Watch this if you want a fast read on whether Meta is back in the frontier race.
Fireship — Every app on Earth is now open source... whether it likes it or not
Fireship tests a developer’s project that rewrote seven Adobe apps in Rust and released them for free, and puts the free versions through their paces. It is a quick, dense look at how far a determined builder can now get against big commercial software. Watch this if you sell software and want to see where the pressure on pricing is coming from.
Nate Herk — I Built the Ultimate AI Board of Advisors (free claude skill)
Nate Herk builds a free Claude skill that works like a personal board of advisors, putting your idea in front of a council of AI voices before you commit. He shows how the skill is set up so you can copy it. Watch this if you make big calls alone and want a sparring partner that pushes back.