Bitcoin Brief

Sup, freaks.

Everybody is focused on what GPT-6 Astra can do inside a browser or a spreadsheet. The bigger question is what happens if governments decide this thing is a strategic weapon.

LEAD STORY

If GPT-6 Astra Is Legit, Governments Will Tell Us

ARC Prize measured Astra at 62.7% at max effort under the same provider-neutral setup that produced 7.8% for GPT-5.6 Sol. That is an impressive jump. OpenAI's own adapter pushed the score much higher by preserving reasoning state and compacting long conversations, though those results are not apples to apples with the standard test. None of this proves that OpenAI has achieved AGI. We have not gotten our hands on Astra yet, and a benchmark is still a benchmark.

But let's assume for a second that the numbers are directionally right. If Astra can enter an unfamiliar environment, work out the rules and execute with far less hand-holding, this becomes much bigger than another round of companies squeezing more productivity out of their employees. That will happen. Some companies will use these systems well, others will light money on fire, and a lot of white-collar work will change quickly. The more important question is whether the capability becomes leverage between states.

Artificial intelligence is now part of the great game between the world's major powers. The country with the strongest systems may be able to move faster across cyber operations, intelligence analysis, weapons development, logistics, industrial planning, scientific research and capital allocation. Models cannot print atoms, build a refinery or manufacture an interceptor by themselves. They can help a state decide what to build, find bottlenecks faster and coordinate the people and machines doing the physical work. If the gap between Astra and competing systems is real, governments will not treat it like a better office-software release.

This is what I will be watching next. Do international competitors dismiss the benchmark, or do they react? Do they accelerate state-backed model programs, tighten controls around chips and compute, secure more energy, deepen alliances with domestic labs or restrict access to their own frontier systems? None of those reactions would prove AGI either. They would tell us that the people responsible for national power believe the capability matters.

We have been circling this point in conversations with John Arnold, Matt Dines and others. AI does not sit in a clean technology bucket while trade, money, energy and war happen somewhere else. It is part of the same macroeconomic and geopolitical reordering. If states are locking down supply chains and trying to rebuild military and industrial capacity, a legitimate step-change in AI becomes another source of leverage inside that contest.

I recorded a conversation with Michael Every, Global Strategist at Rabobank, last night. We got into the shift from market-first globalization toward trade and security blocs, the pressure to rebuild physical capacity, the growing temptation to steer credit, and where stablecoins and bitcoin fit if monetary systems become instruments of state power. The episode publishes tomorrow morning.

Astra fits inside that frame. An AI system cannot create energy, chips, factories or ships on command. The question is whether better intelligence helps a state organize those scarce things faster than its rivals.

Greg Brockman reportedly believes OpenAI may have entered the AGI era. Maybe he is right. Maybe the number is another impressive result that looks less revolutionary once people begin using the model in the wild. I want to test it before making a grand declaration. In the meantime, watch the international reaction. If Astra is legit, the clearest signal may come from the governments that cannot afford to pretend it is not.

SIGNAL

Debt-Heavy Governments Want the Banks to Finance Their Priorities

Governments are "awash in debt", in Scott Bessent's words, but they still want more power generation, manufacturing and infrastructure. The G20 finance track's chair statement calls for private investment, industrial capacity, faster sovereign-debt workouts and common rules for stablecoins and cross-border payments. Politicians want growth without another obvious money-printing blowout.

Matt Dines thinks commercial-bank balance sheets could become the next transmission mechanism, with credit steered toward whatever the state declares strategic. The G20 statement does not adopt his idea, and Kevin Warsh's Jackson Hole speech does not confirm it. He may be right about where this is heading. Reindustrialization may be useful, but if it runs through a more politically managed banking system, the people allocating credit gain even more power over the economy. Bitcoin sits outside that committee, exactly where I want it.

America's Foreign-Born Population Estimate Is Falling

The monthly Current Population Survey estimates that America's foreign-born population fell from 53.3 million in January 2025 to 50.5 million in July 2026. A direct tabulation of the weighted Census microdata puts the decline at 2.858 million. That is a big move, but I want to see it survive the next few releases before calling it a lasting reversal after years of rapid growth.

Foreign-born includes naturalized citizens, lawful residents, temporary visitors and unauthorized immigrants. The estimate can change through arrivals, voluntary emigration, removals and deaths, while sampling, nonresponse and the Census Bureau's January 2026 population-control update add uncertainty to the comparison. CIS infers negative net migration of about 2.45 million after accounting for deaths, while the Census Bureau's latest official projection for July 2025 through June 2026 remained positive at 321,000. Morse Report's path to 38 million by January 2029 is its own extrapolation, not a Census forecast. The next few CPS releases will tell us whether the drop has staying power.

Bitcoin Is Knocking on $82,500 Again

James Check's latest Checkonchain note argues that bitcoin is moving out of the 2025 and 2026 bear market and into the early stretch of a new bull. His line in the sand is the May high around $82,500. Take it and hold it, and the bears need a much better explanation for what is happening.

There is more than price behind the call. Checkonchain's public data show realized cap rising by roughly $8.2 billion over the last 30 days, while MVRV moved back above its one-year average on September 3. Kraken separately shows bitcoin traded as high as $82,288 that day, just shy of its $82,791 May high. Realized cap is a model, not a literal count of new dollars entering bitcoin, and one technical break cannot promise a straight line higher. I still like the setup. After a long summer near the lows, $82,500 looks like the door out of the bear market. We're going to win.

Empire Needs a Story Before It Can Take Your Stuff

Toby Rogers has a useful frame for understanding power. Empire rarely sells conquest as conquest. It wraps the transfer of land, labor, wealth and authority in a story that makes the people taking look righteous and the people resisting look backward or dangerous.

You do not have to agree with every part of Rogers's essay to recognize the trick. Britain defended the Opium Wars with the language of free trade. European powers put a civilizing mission around their conquest of Africa. The details change, but power always wants extraction to look virtuous. I see the same instinct in the fiat system when debasement becomes stimulus, bailouts become stability and financial surveillance becomes safety. Ask who gains control and who pays for it. If the policy cannot survive plain language, the story is doing most of the work.

Zaprite Is Fixing the Awkward Part of Paying Friends in Bitcoin

Zaprite is rolling out early access to a new P2P app that lets users find contacts and send or request bitcoin without first asking which wallet they use, copying an address or passing an invoice through another app. Zaprite P2P is not a wallet. It connects to the wallets people already use and acts as the coordination layer between them.

That sounds simple because it should be simple. The fact that it has taken this long shows how fragmented the bitcoin payment experience still is. Zaprite says contact matching uses phone numbers that are hashed and salted on the device, with no contact list sent to its servers. The app is in limited early access on iOS and Android, and we have not tested that privacy design or the payment flow ourselves. If it works as advertised, this is a meaningful UX improvement. Bitcoin payments should feel like sending a message, without forcing everyone onto the same custodial platform.

ETF FLOW CHECK

U.S. spot bitcoin ETFs recorded an estimated $730.9 million net inflow on September 3, according to TFTC's ETF tracker. IBIT led with $454.0 million, followed by ARKB at $137.7 million and FBTC at $74.4 million. HODL and BTCW posted combined outflows of $24.7 million.

The latest session extended the inflow streak to two trading days and $832.0 million. The funds have taken in a net $595.6 million through the first three sessions of September. Two days are not a trend, and IBIT did most of the work. If the bid starts spreading across more of the complex, we will have something stronger than one BlackRock-led burst.

⚡ FREEDOM TECH CORNER

Nunchuk Android Added BitBox and Krux Support

Nunchuk Android 2.8.5 adds support for BitBox over Bluetooth and USB, adds Krux support, and introduces emergency push notifications and home-screen alerts. Nunchuk describes the app as a native Android multisig wallet powered by libnunchuk and publishes reproducible-build instructions.

The release also includes a direct APK, a checksum file and a signature for the checksum, so users can verify what they downloaded outside an app store. BitBox and Krux support widen the set of hardware that can be used in a mobile multisig setup. The notes do not say which BitBox models or multisig paths were tested, or what triggers an emergency alert. Test recovery and signing with a small amount before trusting a new device pairing.

DATA SNAPSHOT

As of September 4, 2026, approximately 10:17 a.m. ET

Sources: Kraken for spot price; mempool.space for block, fee, hashrate, and difficulty data. Bitcoin Lab's on-chain client was unavailable at capture time, so no unavailable model metrics were backfilled or reconstructed.

Stay humble. Stack sats.

YouTube: https://www.youtube.com/@TFTC

podcast: https://www.tftc.io/tag/podcasts/