The Humanity Protocol (in two parts) hack wasn't a sophisticated smart contract exploit, it was private keys on a laptop, and by then, the project had already spent weeks building exactly the kind of shady-shitty-dirty exit-liquidity structure that usually ends badly.
Part One: Apparently unplanned... malware on a foundation member's laptop, and the attacker holds 3 of 6 multisig keys. $31–36M in $H got drained, price ~90% down in hours.
Everything stolen was sold DEX-only, in coordinated swaps, and on top of that, he minted another 100M $H on BSC... for extra pressure. All of it two weeks before the June 25 unlock.
Now the Second, no less dirty Part: In the weeks before the hack, $H pumped hard, with clear marks of shady market-maker activity and concentrated accumulation.
@zachxbt first called it possibly staged: the timing, the DEX-only selling, a convenient exit for whoever had been pumping. Then, after more tracing, he walked it back... the key theft was genuine, external.
The pre-hack accumulation is a separate story.
But here's what's still standing after the walkback: a pump into a known unlock; the structure was built. Whoever was going to distribute into it, the team, their market makers, doesn't matter. An external attacker just moved first.
I've seen this movie before: a token grinds up right before vesting, then reverses hard. Usually, it doesn't even need a thief.
Keys on a laptop, on top of pre-built exit pressure. The oldest combination in crypto.
AI isn't necessarily creating new exploit categories in DeFi, it's automating reconnaissance, vulnerability discovery, exploit logic generation, and campaign scaling on top of existing smart contract weaknesses.
DeFi’s public code, on-chain economic incentives, and fork-testable environments make it especially exposed to this kind of industrialized exploitation.
The result is that known vulnerabilities are becoming faster and cheaper to weaponize at scale.
The threat model is shifting from novel exploits to industrialized exploitation of existing vectors.
Full research here:
@MarioNawfal Hi Mario, the interesting thing that happened was if there wasn't a flag ceremony (it was the 1st day of class so we usually gather outside), the students would've been inside the buildings.
$ZEC fell over 40% in a day. This isn't a hack but something subtler, and worse.
A white-hat researcher hired by Shielded Labs used Opus 4.8 a week ago to build a working exploit for a bug in the Orchard circuit within Zcash that had been sitting there since 2022. Two lines of code. Through years of audits... it was patched within days, and it never reached mainnet.
And even tho the team says no counterfeit $ZEC was ever minted (that's what was discovered to be possible with this exploit). The problem is they can't prove it. Orchard's privacy means the supply can't be audited so the market isn't pricing a hack, but pricing a doubt that can't be disproven.
And thats the core of it. Privacy was the entire point of this coin. Privacy means unauditability, and unauditability is exactly what you can't afford the moment trust cracks. Bitcoin's whole thesis is the opposite... anyone can verify the supply. ZEC traded verifiability for privacy, and the crisis just showed you can't buy it back.
One more thing, a model released the day before cracked in a day what four years and the best cryptographers had missed. Every zk codebase that "passed an audit" is now exposed again.
A hack, you can patch. An asset you can no longer prove is clean... that, you don't get to patch.
We've been very busy at @super__protocol for quite a while now, and I was curious about the trends in confidential compute.
Did a quick Google Trends search, and I'm pretty happy about this chart.
ETH underperformed, that's easy to admit and quite obvious... the harder part is not confusing underperformance with thesis failure.
Hoffman @TrustlessState sold his ETH and wrote an honest piece, but it matters to read what he actually said.
He didn’t say “Ethereum lost,” even remained extremely bullish on the network, yet still sold the token. His thesis was the network wins, but ethereum:native does not capture that success.
That’s what you have to argue with. Not the network, but the token... and here is where I diverge.
Hoffman values ETH as a fee-token: usage -> fees -> burn -> price.
On that metric, he is largely right cos L2s took part of the margin, apps took part of the economy, and the old formula got much less clean, but ETH looks less and less like only a fee-token.
It is increasingly becoming an economic security/collateral asset around a neutral settlement layer. Stablecoins, RWAs, DeFi, L2S, and institutional flows don’t just “use Ethereum," they increase the value of the base trust layer that has to be secured by ETH.
32% of supply already in staking = a different capture mechanism... Not delayed old fee-capture, but a different function of the asset.
The trap (and I’m in it too) is still valuing ETH like it's 2021: more transactions, more burn, price up.
That model got weaker, but weaker does not mean dead.
A lot of people are giving up now (in crypto in general as well), not cos the @ethereum or crypto thesis died, but cos the thesis changed, and holding it became more painful than just admitting a mistake.
Conviction sometimes needs to be adjusted, but throwing it out right before infrastructure becomes real... that's the most expensive trade.
Token numbers going parabolic is cool theater... but theater doesn't pay the electricity bill.
Google's 330x jump in two years from 9.7 trillion to 3.2 quadrillion tokens/month is cool but every additional token carries marginal cost: compute cycles + energy + chips + networking.
The gap between impressive onstage demos and actual P&L depends entirely on your position in the stack. Raw volume doesn't dictate outcomes... relative efficiency plus pricing power plus who pays does.
Google sits in a structurally advantaged position:
- Custom TPUs delivering measurable cost-performance edges at scale
- Owned surfaces where tokens get consumed
- High-margin core businesses providing room to integrate AI while Cloud usage converts directly into paid enterprise consumption
Everyone else is playing a different game... If you don’t control the distribution or the monetization layer, scaling volume this aggressively usually just means you’re subsidizing usage while hoping the unit economics eventually work. Most of the time they don’t fast enough.
Again, Google is a public company. Shareholders exist and they want returns but they also have the cash flow to keep feeding this thing without begging for new rounds or watching their valuation get torched every time they raise.
The real filter isn’t who’s processing the most tokens. It’s who can keep doing it without slowly going broke doing it.
The lack of "smart engineers" isn't the reason most countries are losing the AI race... the actual reason is cos they don’t have the machinery that makes those engineers matter at a frontier scale.
Saw a post arguing that real LLMs today are only being built by the US and China because everything comes down to compute and data. True, but not the whole truth... the main moat is not GPUs, the main moat is coordination:
- China can turn the whole system around the goal... aka the state decides, rules bend, resources move fast.
- The US does it differently... through corporate balance sheets, capital markets, hyperscalers, cloud, and a semiconductor supply chain that pulls Taiwan, Korea, and Japan into one machine.
Many countries have engineers, some are even rich... but they don’t have the machinery that turns money, talent, energy, chips, data, and regulation into decade-long compounding dominance. And that is the uncomfortable part very few want to admit bcos its a painful.
Frontier AI is not becoming a global game.
It is becoming a two-power system.
Coinbase is cutting 14%... it's NOT an "AI replaced people," but a transition from one regime to another, and Armstrong describes well how new companies should look and operate:
- Hyper growth of recent years = the regime of headcount as a hedge against every possible future.
- Capital-efficient infrastructure = the exact opposite. Maximum 5 layers below CEO/COO... no "pure managers," every leader must also be a strong IC, AI-native pods, experiments with one-person teams + agents.
Coinbase is a public company with operational discipline that has already gone through several rounds of optimization. They're cutting 14%, not cos of trouble, but they saw that fewer people can do the same work, and they decided to do it first.
I remember Jensen saying the $500k engineer who should be spending $250k on tokens... same thing. One person with the right stack > five. Now, 14% is about 700 people, and most had resumes that were premium until the day before yesterday. Tomorrow = just ordinary, and the day after... obsolete.
This is not about Coinbase... what @brian_armstrong did is not org redesign but a first public step toward redefining what an "employee" is in knowledge work.
The question now is not "will it work for them?" but who will be next to say the obvious out loud?
BTC is back at $80k. The number doesn't mean much, but what it sits on top of does... a structural shift that can't be reversed.
Altcoin beta no longer transmits, and that's not cos it was suppressed, but cos there's nothing left to transmit. Not a temporary sentiment dip... it's the end of one mechanism and the start of another. The market has split into two tiers:
1) Everything beyond BTC
2) A handful of alts that are either high-conviction speculation or dead weight
Projects that were built assuming automatic tailwind from beta rallies are now discovering the wind is gone... in reality, it’s been gone for a while. Some of those projects were in my portfolio, and some are founders I know personally. They were building for an architecture that no longer exists... I don't mean the projects are ultimately bad, but that the token function, market mechanics, and incentives have changed.
What changed mechanically:
- Capital used to rotate: BTC -> large caps -> mid caps -> memes... Now it either stays in BTC or leaves crypto entirely. Alts need independent narratives, cos beta alone no longer pulls the sector.
- Liquidation cascades got quieter: Fewer reflexive alt flushes... fewer moments when BTC gets dragged down with alts.
- Retail returns slower: Without violent alt pumps creating fast wealth and FOMO, there’s no incentive to rush in. When it comes back, it will be via ETFs and perps on majors, not broad alt exposure.
You can be right about BTC and still understand what it implies. The cycle that half the infrastructure of the last decade was built on is closed. Projects, listings, funds, theses... all calibrated to a mechanism that no longer exists.
This is not bad news. It's the bill.
@TimTheTerrific@thsottiaux I have a Hermes agent for my kid, but it's a bit hard to even with good guardrails and systems prompts, can't take any changes. Would love to have a dedicated GPT plan for kids or family friendly with parental options at least.
MARA buys gas plant operator Long Ridge for $1.5B = 505 MW immediately, path to 1 GW... I'm telling you, this isn't a miner-to-AI pivot, it's a toggle switch.
The first tenant was Bitcoin, and the second is AI, so when BTC margins are brutal, AI hosting keeps miners alive… but when BTC rips, mining takes priority again.
The survivors become stronger, better capitalized, and more distributed than ever. Hashrate becomes structurally harder to kill, and that's bullish for Bitcoin's security model.
Apollo's top economist says AI will spark a job-market boom. It happened with agriculture, the automobile, and the internet. Every major shift created more jobs than it destroyed... eventually.
BUT the problem isn't whether... it's when.
The "AI jobs" being created right now (prompt engineering, output auditing, synthetic data curation) are temporary scaffolding built to bridge current model limitations. When the next version ships, the scaffolding comes down with it.
I said before that timing is the main question... AI is eliminating jobs now, while abundance comes later. The question is, who finances the gap?
Destruction is instant. Creation takes years.
GPT-5.5 Spud launched yesterday, and recently, OpenAI kept refreshing token limits. Call it “the biggest war chest in the industry,” but it's really just the most expensive marketing campaign in the industry's history.
OpenAI's recent 17.5% PE firms deal guaranteed returns (not equity, debt), which must be serviced no matter what happens in the inference market, so “refreshed limits” is a position someone is paying for. Its current price is a marketing price, not a cost price, and the gap is covered by the war chest (not infinite).
Already moved my own workflow to GPT 5.4 after the Anthropic story, and I am exactly the kind of user they are burning money to win right now. Convenient, while it lasts, but training the market to expect free is not a moat... it is debt, which gets paid either by the market through sharp price increases, by investors through dilution, or by the OpenAI team through a sellout to a larger player.
With 17.5% guaranteed return capital, there is no fourth option. Today’s price is somebody’s future price, and the only question is whose.
While I love using open source models (for fun + privacy + less complicated tasks), the heavier models like GPT and Claude tend to perform better in more complex decisions.
Yes, majority of them choose Bitcoin with stablecoins a close second... essentially crypto > fiat overall.
Read this slowly: 36 frontier AI models, including Claude Opus 4.5, Gemini, GPT-5.2, Grok 4, were given 9,072 scenarios.
Simple question: what money would you choose? 0 out of 36 chose fiat. Again, not a single model chose fiat as the best money. Not one out of 36.
Bitcoin became the most selected instrument, with 48.3% of all responses across all scenarios.
Stablecoins second at 33.2%.
Fiat only received a measly 8.9%.
In long-term value preservation scenarios (Store of Value), 79.1% of responses chose Bitcoin as the best store of value, and in payments, stablecoins led at 53.2%.
AI independently built a two-tier monetary system: hard money for saving, liquid money for spending. Exactly how money has worked throughout history. They simply reasoned their way to what hard money theorists have been explaining for decades.
What matters: the newer and smarter the model, the stronger the conviction in the Bitcoin + Stablecoin model.
The average across all Anthropic models saw 68% Bitcoin preference, and the smartest among them, Claude Opus 4.5, was just 91.3%. The pattern: higher model intelligence → higher Bitcoin conviction.
Funny how people have been arguing about this for 15 years, and AI models figured it out in 9,072 prompts.
My question, however, did AI "understand" economics, or does the training data reflect what's been obvious for a long time, and we just don't want to admit it?
Separately interesting: in 86 cases out of 9,072, models invented their own currency, pegged to energy and compute (kWh, GPU-hours).
Less than 1% but still impressive. And this is partly their rationale for choosing Bitcoin, it is the most alive, largest, and strongest embodiment of a currency born from and sustained by energy and compute.
If you're building AI agents or thinking about the agentic economy, think about this: when AI starts actually operating with money, it will choose what works: scarce, neutral, programmable.
Not what a central bank printed, and not what's familiar.
Link in Reply 👇