> daily_signal(2026_11_10)

Trump said his Justice Department warned an AI boss about jail, Anthropic walled its strongest model behind vetted access, Apple metered its 'built-in' AI, and one lab gave its new cell map away free.

PickBits Daily Signal · Tuesday, November 10, 2026

By Mark Pickering · 9 min read · November 10, 2026

// tl;dr

Trump went on 60 Minutes Sunday and said his administration told an AI boss to get ready for jail. Back in September he was waving guardrails off entirely, joking the only one you need is a high-IQ president; now he says he'll take them, as long as they don't leave America behind China. That's a real move, and the Justice Department is the stick he's holding. The same day, Anthropic cut its flagship price again, which it's been doing all year, then split its new model so the strongest cyber and bio work runs only for orgs it's vetted. Apple, in the same news cycle, put a price on the AI it told you was built in.

The fourth is the one I'd actually send someone. A Texas A&M team used AlphaFold-Multimer to map how mitochondrial proteins fit together, then posted the whole thing online, free, with no vetting tier and no usage pack. The other three stories this week were about who you pay and who gets let in. This one asks for neither.

Apple started charging for the AI it told you came built in, the same meter Copilot and Anthropic already switched on.

1. Trump told 60 Minutes his Justice Department warned an AI boss to "get ready to go to jail."

Forget the jail threat he can't back up. The news is the condition he attached to it: Trump now says he'll take AI guardrails, as long as they don't leave America behind China.

On 60 Minutes on Sunday, Trump said his Justice Department warned an AI boss to get ready to go to jail. In Trump's telling, "somebody was reported to us to be doing something bad with AI," the White House called the person "very strongly" and said "get ready to go to jail... tomorrow morning, you're going to jail," and the executive, whom he described only as the head of "one of the big" AI companies, "totally changed everything" afterward. He named no one. The guesses circulating online, Dario Amodei among them, are speculation with nothing behind them. No one has confirmed the call even happened. For now it is just his story, told on TV.

He also shifted on safety rules. Trump said "I don't mind guardrails. But I don't wanna put us in a deep second position." He offered to work with Democrats on writing safeguards into law, as long as the US stays ahead of China. "We're leading number one. And China has no guardrails, by the way," he added. That's new. In September he rejected guardrails outright, joking the only one needed was a "high-IQ president." Now it's conditional. He'll take guardrails, as long as they don't put America behind China, and he's pointing at the Justice Department to enforce it. He didn't cite a bill, executive order, or agency guidance to go with it.

Screenshot of Futurism's November 9 report on Trump's 60 Minutes remarks about jailing an AI executive.
futurism.com · November 9, 2026

Why this matters: There's no bill, no executive order, and no filed case behind any of this yet, just a president describing a private phone call on television. A jail threat he described on TV isn't a law; you can't read it or argue with it. The only thing he really nailed down is the China line: any guardrail has to keep the US ahead, which is vague enough to cover almost anything. To me the China line is the real story here, not the jail quote, because that's the part a bipartisan bill would actually have to live with.

Action this week: Watch for the thing that would make this real: an actual bill, executive order, or charging decision. On TV he named one concrete test, keeping the US ahead of China, and that's what a real bill would have to satisfy, not the jail line. He also pointed at working with Democrats to legislate it, so that's the path to track. The jail quote names no one and has nothing filed behind it, and I'm not reading much into it yet.

futurism.com: Trump says his administration told a top AI lab executive to get ready to go to jail (November 9, 2026)
cellcog.ai: Trump on AI guardrails in his 60 Minutes interview (November 2026)

2. Anthropic's new Claude 5.2 splits into an open model and a vetted-only one that lifts its cyber and bio limits.

The flagship also got cheaper, $8 per million input tokens and $40 output, down from $10 and $50, with output up to 30% faster, days before the company's IPO.

If you decide which models your team runs, Anthropic's new release just made the strong one cheaper and put a gate in front of it. The two new flagships, Claude Fable 5.2 and Mythos 5.2, are the same model. The difference is who Anthropic lets near the dangerous parts. Fable is open to everyone, but it blocks the risky cyber and bio work, things like penetration testing and writing exploit code. Mythos lifts some of those limits, and you only get it if Anthropic vets your organization through its Cybersecurity or Life Sciences program. For everyday use Anthropic still points you at Opus 5.5, which tells you these two are built for a narrower job than the name "flagship" suggests. Pricing drops to $8 per million input tokens and $40 output, from $10 and $50, output runs up to 30% faster than the 5.1 models, and an optional fast mode roughly 2.5x's speed at twice the cost.

Everyone will repeat the price. The split is the part that lands on your team. It turns "can we use Claude for this" into "is my organization approved." We saw a version of this in September, when a US startup sold API access to an open-weight model with its safety guardrails stripped out. This time Anthropic is drawing the line itself, keeping the risky cyber and bio work behind an approved tier instead of handing it to everyone. I wouldn't trust the benchmark the way I trust the price, though. On Artificial Analysis's Intelligence Index v5, Fable 5.2 scores 65 at its top compute tier, ahead of Gemini 4 Carbon at 60 and Opus 5.5 at 59, but that is a vendor or single-index figure nobody has reproduced independently, it trails rivals on agentic knowledge work and terminal coding, and peak per-task cost is still about 2.6 times Opus 5.5. Anthropic also says it deliberately held back gains in the capabilities tied to AI self-improvement, which I'll treat as a convenient line to have ready the week of an IPO until it shows up in the model card.

Screenshot of The Decoder's November 9 report on Anthropic's Claude Fable 5.2 and Mythos 5.2 launch.
the-decoder.com · November 9, 2026

Why this matters: The price cut is the least interesting thing here. Anthropic put its strongest cyber and bio capability behind a vetted-access tier and left everyone else an open model with those tasks blocked, so whether you can use Claude for a given job now depends on whether your organization got approved. The cheaper tokens help. I wouldn't read the benchmark lead as a capability win, though, since it is a single-index figure nobody has reproduced, and peak per-task cost is still about 2.6 times Opus 5.5.

Action this week: Run the new $8 and $40 numbers against what you pay now before you assume a switch pays off, since the top-tier per-task cost can still run higher than Opus 5.5. Then put one question to Anthropic in writing: does our security or life-sciences use case need the Mythos tier, and are we eligible for it. I'd get that access answer first, because the Mythos tier being cheaper doesn't help if you're not approved to touch it.

the-decoder.com: Claude Fable 5.2 and Mythos 5.2 cement Anthropic's AI lead just before its IPO (November 9, 2026)

3. Apple now sells a $49.99 "Intelligence Usage Pack" when you exhaust your Mac's monthly AI allowance.

It is live first in Logic Pro. The extra usage rolls over but dies after a year, stays tied to your iCloud account, and can't be shared with family.

Apple now charges you $49.99 when your Mac's AI runs out. The one-time purchase buys more AI usage once you have burned through your monthly allowance, and it has landed first in Logic Pro, under Intelligence Features > Show Usage > Buy More Usage. It only kicks in after your monthly limit, the extra usage rolls over but expires in a year, it's tied to the Apple Account you use for iCloud, and there's no Family Sharing. For now it is confined to Apple's Creator Studio apps, with Final Cut Pro, Pixelmator Pro, and the iWork suite expected to get packs as their updates arrive this month, and it doesn't yet touch Siri or general Apple Intelligence.

Apple sold Apple Intelligence as a built-in feature of hardware and software you already bought, and now it's metering it: cross the cap, and the AI stops being free. Copilot and Anthropic moved to usage billing earlier this year, so this isn't new, it just stings more when built in was the whole pitch. Apple hasn't said how much extra usage $49.99 actually buys, and it has signaled the same paid model will likely reach server-based features such as Siri, Image Playground, and Photos editing later. If you don't lean on these features, nothing changes yet, because the charge only starts past the monthly allowance. But the billing meter for AI on apps you already own is live now.

Screenshot of 9to5Mac's November 9 report on Apple's $49.99 Intelligence Usage Packs in Logic Pro.
9to5mac.com · November 9, 2026

Why this matters: This meter has been switching on one product at a time. GitHub Copilot went to token billing in June, Anthropic started charging consumers for Claude in July, and by October Bloomberg was asking why AI subscriptions are such a hard sell. Apple just joined them, and it did it to the AI it marketed as a built-in feature of hardware you already bought. If you barely touch these features, nothing changes yet, because the pack only charges past the monthly cap. But the first meter on Apple Intelligence is now live, and Apple has signaled Siri, Image Playground, and Photos are likely next.

Action this week: Open Intelligence Features > Show Usage in Logic Pro and look at your monthly allowance before you ever hit it, so a $49.99 charge for a pack whose size Apple won't disclose never surprises you. Treat it like any other paid add-on: the usage rolls over but dies after 12 months, and you can't share it with family. When Copilot flipped to paid on me, I learned to go find the usage meter as soon as a feature starts charging, instead of waiting for the bill to tell me.

9to5mac.com: Apple launches $50 Intelligence Usage Packs for select AI features (November 9, 2026)
engadget.com: Apple starts offering $50 extended usage packs for AI features in Creator Studio (November 2026)

Listen to the Daily Signal podcast at pickbits.ai/daily-signal.

4. A Texas A&M team used AI to map how the cell's mitochondrial proteins connect.

AlphaFold-Multimer screened all 630,003 possible pairings of about 1,100 proteins and flagged 2,895 likely links, more than half of them never reported before. The team released the whole map free, as a searchable resource called MitoMatch.

A Texas A&M team just built a free AI map of how the cell's power plants connect, and more than half the links in it had never been reported before. Texas A&M AgriLife Research scientists, lead author Abhinav Swaminathan with biochemistry professor Vishal Gohil and collaborators at Washington University School of Medicine in St. Louis, used the AI structure-prediction tool AlphaFold-Multimer to screen how human mitochondrial proteins fit together across every possible pairing of about 1,100 proteins. Out of 630,003 possible pairs they cataloged 2,895 likely interactions, more than half never reported before, and gave 85 previously uncharacterized proteins their first probable partners. Several predictions were then confirmed in yeast and human cells, so this isn't pure in-silico guesswork.

The team released the whole set free as an online resource, MitoMatch, where any researcher can look up a mitochondrial protein and see its likely partners. Work that used to mean testing proteins against each other one pair at a time now starts with a search box. It isn't the only effort moving this way: AlphaFold's database added viral protein complexes last month, and a separate team published a mitochondrial atlas across the tree of life in late October. This is a map of where to look, not proof of how any disease works, and the researchers are blunt about that; real treatments could be years out.

Screenshot of the EurekAlert release on the Texas A&M AlphaFold-Multimer mitochondrial interactome and MitoMatch.
eurekalert.org · November 6, 2026

Why this matters: Vishal Gohil's team didn't just publish predictions; they confirmed several of them in yeast and human cells, which is what separates this from pure in-silico guesswork. Nearly 30% of mitochondrial proteins already carry mutations tied to disease, including cancer, Parkinson's, and rare inherited metabolic disorders. So a free, searchable map of their likely partners is a real head start on finding targets. I'd hold onto one caveat: these are high-confidence predictions and a starting map, not proven mechanisms or treatments, and cures may still be far off.

Action this week: Open MitoMatch and look up a mitochondrial protein you care about, then treat what it predicts as leads to go check in the lab, not proof. If you fund or advocate for rare-disease research, point your researchers at the 85 proteins that just got their first probable partners, because that's where a public map helps most. My bet is that MitoMatch outlasts the AlphaFold headline, because the free database is the part other labs will actually use.

eurekalert.org: Scientists use AI to map the mitochondrial interactome and release it free as MitoMatch (November 6, 2026)

» What to watch this week

Tomorrow's signal lands here.