The Record

Block No. 11 - The Record

Five brands start a 12-week sprint, and the first thing I’m going to ask them for is their co-packer agreement. The AI labs asked Washington for permission to slow down together, then shipped four frontier models in three weeks. Three different ways of holding bitcoin got hit in eight weeks while the chain wrote down every transfer. And I built a screen that never stops updating, with one number on it I can’t stop looking at.

Everything leaves a record.

Business - Nobody Fails on the Shelf

This fall I’m mentoring in Track 7 of BeyondCPG, BeyondBrands’ accelerator for emerging consumer brands and its first nationwide cohort. Five founders, 12 weeks, and a January showcase for investors and retailers.

Every founder of a food or beverage brand is already staring at the product risk. Will it sell through, will the buyer take the meeting. The risks that kill brands usually live in the paper.

Start with the co-packer. Most early brands don’t make their own product, and the manufacturing agreement is usually the manufacturer’s form, signed the week the first run got scheduled. Read it a year later and you find the minimums, the price escalators, a clause giving the co-packer ownership of process improvements it made on your formula, and a termination provision that leaves you without a line for four months.

Then the retailer. The purchase order feels like the win. The vendor agreement behind it carries slotting fees, promotional commitments, chargebacks for a late truck or a mislabeled pallet, and deductions taken straight out of what they owe you. A brand can grow revenue and lose money on every unit, and the P&L won’t tell you why until somebody reads the deduction report line by line.

I spent nearly seven years in-house at a venture-backed e-commerce and logistics company selling consumer products, through a $2B+ SPAC deal and a wind down. The founder who knows their contracts cold negotiates from strength with every counterparty. The one who doesn’t renegotiates from weakness, after the damage shows up.

These five founders have the product handled. My job is the part that never makes it to the shelf.

AI - Permission to Slow Down

On September 12, Dario Amodei published a proposal for the labs to pace the frontier together, with the government granting a narrow antitrust waiver so rivals could have safety conversations without breaking the law. Embedded evaluators would verify it. David Sacks, who co-chairs the President’s Council of Advisors on Science and Technology with Michael Kratsios, answered that weekend: no waiver. If you want to slow down, go slow down. He accused them of trying to form a cartel.

Six days later, four subscribers sued Anthropic, OpenAI, SpaceXAI (formerly xAI) and Google in the Northern District of California. Buist v. Anthropic alleges a horizontal agreement to restrict output, which is the plainest Sherman Act Section 1 theory there is. The evidence is entirely public. No leaked memos. Just the labs’ own essays and posts and, as the complaint cites, OpenAI policy chief Chris Lehane’s public confirmation that they’d been working with Anthropic and Google DeepMind for weeks.

Anthropic shipped Fable and Mythos 5.1 on September 1. OpenAI answered with GPT-6 Astra on the 3rd. SpaceXAI put out Grok 4.7 on the 21st. Claude Opus 5.5 landed on the 22nd, performing at Fable’s level on most work for 40% less to run than Opus 5.

Four frontier releases in three weeks, from the companies asking for permission to slow down.

Here’s the legal problem. There’s no waiver an adviser can hand out. Antitrust exemptions come from Congress, and the closest thing the agencies offer is a business review letter that binds the Justice Department and nobody else. Private plaintiffs don’t need anyone’s permission to sue. They just did.

Unilateral restraint is completely legal. What they need a waiver for is agreeing with each other, which is the one thing the Sherman Act was written to stop.

I use these tools every day, and I want the people building them to be careful. But I’ve seen enough to know what it looks like when competitors want to talk to each other about output and want immunity for the conversation.

Slow down or don’t. Just don’t ask permission to do it together.

Bitcoin - Three Hacks, One Chain

Between July 30 and September 24, three different ways of holding bitcoin got hit.

Coldcard first. A firmware bug dating to March 2021 left some devices generating seeds from weak software randomness instead of their hardware entropy source. On older units, key strength dropped from 128 bits to as little as 40, low enough to brute-force from anywhere. TRM Labs counts roughly 1,816 BTC taken from more than 5,200 addresses in the first week. That’s self-custody. The keys were yours. They just weren’t secret.

Then Liquid, Blockstream’s federated sidechain, on September 6 and 7. An attacker pulled about 4,000 of the roughly 4,200 bitcoin in the federation wallet, called it white-hat work, and gave back 3,400 the next day. They kept around $47 million as a fee they set themselves. That’s federated custody: you trust a group of functionaries and the software that ties them together.

Then Bitget on September 24. About $387.5 million out of hot and warm wallets by the exchange’s updated count, with XRP leading a mix spread across at least five chains. Withdrawals frozen. The exchange says its $464 million protection fund covers the loss. That’s custodial: you trust a company and its balance sheet.

Not one of those failures was Bitcoin. The chain kept producing a block roughly every 10 minutes and recorded every one of those transfers exactly as instructed.

Not your keys, not your coins still holds. Coldcard adds the fine print: your keys, your problem. Self-custody moves the risk onto you, meaning the hardware, the firmware, the seed ceremony and where the backup lives. Every way of holding bitcoin carries its own risk.

Pick yours knowing that.

IRC - The Life and Times

I built a live dashboard for myself that sits open all day and tells me what’s true right now.

Markets, Bitcoin, what’s trending. The national debt, sitting just over $40 trillion and running like a taxi meter. The 10-year, gas prices, the Fed funds rate, the Long Island headlines, and whether the Yankees won.

I got tired of rebuilding the same picture from 12 tabs every morning, so I built it with the same AI tools from the section above and left it on.

It refreshes whether anyone’s looking. It’s the thing itself, writing itself down.

The lines that stops me are national debt and housing. National debt is over $40 trillion and climbing rapidly. Median asking price for a home in August: $425,000 nationally. Long Island, around $900,000. And this week the 30-year fixed topped 7% for the first time in 20 months.

Long Island asks more than twice what the country asks.

Three kids who are going to want to live where they grew up, and a number that says it will require a herculean effort. There’s no take here. It’s just on the record now, updating, whether I’ve got an answer for it or not.

You don’t get to choose whether there’s a record. You only get to choose what’s in it.

The Links

Worth your time:

• BeyondCPG: The accelerator behind Track 7 and its first nationwide cohort.

• Buist v. Anthropic, explained: The Sherman Act suit built on the labs’ own public statements.

• Inside the Coldcard exploit: TRM Labs on how 128 bits became 40.

• The Bitget breach: $387.5 million, a withdrawal freeze, and a $464 million fund.

• The IRC Legal Live Business Briefing: Markets, Bitcoin, macro, housing and the headlines, refreshing all day.

I’ll catch you at the next block.

-Ian R. Cohen

Founder and Principal Attorney - IRC Legal

Get the IRC Legal Business Brief in your inbox. Every issue covers business, AI, Bitcoin, and life. Subscribe here →

Block No. 11 | Originally Published Block No. 969153