There's a line buried in an old engineering résumé that has been ricocheting through crypto back-channels for weeks, and almost nobody is reading it correctly. It says that before this person became the head of alignment stress-testing at Anthropic — before the interviews, before the 30-million-view argument about whether superintelligence could end us — he was a software engineer who helped build rippled, the server software that runs the XRP Ledger, and who contributed to Interledger, an open protocol for moving value between ledgers that don't have to trust one another. The internet did what the internet does. "The AI doomer used to work at Ripple," went the joke, followed instantly by the ritual reminder that XRP still hasn't reclaimed its January 2018 high. It was funny for a day. Then it kept spreading, and the laughter began to function as a way of not thinking — a safe noise to make so that nobody has to ask the uncomfortable question hiding in that résumé. Which is this: why is it always the trust people who leave first?
The person is Evan Hubinger. He now leads the alignment stress-testing team at Anthropic, an outfit whose entire job is to construct scenarios in which a model pursues goals we never specified — and to find the failure before it finds us. In the original interview that lit the fuse, he attached a number to that anxiety: a probability he placed above ten percent that advanced AI could extinguish humanity within a decade. He later softened it, insisting that today's models are low-risk and that his true concern is recursive self-improvement — an AI that trains the next, stronger AI without a human in the loop. That clarification mattered far less to the discourse than the headline had, which tells you something about what the discourse was actually for.
Ripple is one of the oldest companies in this industry, and Interledger was among the first serious attempts at cross-ledger interoperability — a way to route value across systems that had no reason to trust each other. rippled is the client that keeps the XRP Ledger alive. Note the governance detail the jokes skip: Interledger no longer belongs to Ripple; it's stewarded by an independent foundation. That's the kind of quiet institutional handoff that matters to a protocol's long-term credibility and matters not at all to a meme.
Meanwhile the price of XRP sits below where it stood in January 2018, when it touched roughly $3.40 — a fact Galaxy Research's Alex Thorn invoked sideways, letting the comedy do the argument for him. And Ripple, per reporting, has been seeking a valuation of as much as fifty billion dollars, a figure underwritten in large part by the company's own XRP holdings. Hold those two facts next to each other — a token below its eight-year-old high, a company valued well above the token's market mood — and a structure starts to show through. More on that shortly.
But the joke is not the story. The story is what a résumé reveals when you stop laughing at it.
Here is the first thing I want on the record. I came into this industry through an audit, not a whitepaper. In 2018 I spent three months — unpaid, obsessive, mostly at night — going through the smart contracts of a fledgling DeFi prototype called EtherTrust. I found a reentrancy vulnerability in their donation logic, the kind of flaw that turns a charitable function into an open drain, and the fix prevented what the team estimated at a two-hundred-thousand-dollar loss. What stayed with me wasn't the bug. It was the moment I understood that a smart contract is a moral document written in a language that has no word for mercy.
Interledger and alignment are the same sentence with different subjects. Interledger asks: how do two parties exchange value when neither will trust the other's intermediary? Alignment asks: how do we get an intelligence to pursue what we actually meant when we cannot inspect its intentions? Both are adversarial trust problems. Both assume the counterparty may be indifferent, self-interested, or actively hostile. Hubinger's career didn't switch fields. It walked the same problem from money to mind. That is not a sentimental parallel — it is a structural one, and it explains the migration better than any price chart.
Crypto's best engineers spent a decade building machinery to remove the need for trust: multisig, time locks, fraud proofs, optimistic rollups, an entire civilizational toolkit for making betrayal expensive. And somewhere in that decade, the sharpest of them looked up and noticed that the trust problem they had not solved was the one inside their own heads — or rather, inside the heads they were about to build. You can make a payment trustless. You cannot, so far, make a mind trustworthy by construction. The people who cared most about trustlessness became the ones most alarmed by an intelligence that could not be audited. The résumé is not a punchline. It is a pipeline.
Which brings me to the second thread, the one the bear market makes urgent. Everyone is treating this as a story about AI. It is also a story about who is bleeding.
Look at the names that surface around the Ripple history here. Hubinger, gone to Anthropic. David Schwartz, the long-serving CTO, now retired and weighing in publicly with the libertarian instinct this community knows well — the sense that AI-safety rules are themselves a threat, that the cure may strangle the speech it claims to protect. Emi Yoshikawa, a former vice president, elsewhere now. Chris Larsen, a co-founder, attached to a murkier thread — the suggestion that nonprofit structures were used to buoy a for-profit XRP reserve. Draw the line through all of them and you get something a single data point cannot give you: not one departure, but a direction. The people who built Ripple's trust machinery are, one by one, somewhere else.
In a bear market, this is the signal that matters and the metric nobody tracks. We measure TVL, we measure active addresses, we measure fee revenue, and we ignore the only leading indicator that has ever predicted a protocol's slow death: whether the people who understand it most deeply still choose to spend their finite attention on it. Attention is the scarce asset now. Everything else — the token, the chain, the treasury — is downstream of whether serious people keep showing up.
Here's a structural detail worth naming, because it's the part that actually threatens the holders. Ripple's corporate valuation rests heavily on its own XRP balance sheet. That welds the company and the token into a single feedback loop: if the token bleeds, the balance sheet bleeds, the valuation weakens, the fundraising cools, the ecosystem investment thins, and the token feels it. Nobody should pretend that loop is a tokenomics question. It is a survival question, and it's the kind of coupling that looks fine until the exact moment it isn't.
I learned the shape of that loop the hard way. During the DeFi summer of 2020 I was a junior community liaison at a lending protocol called LendPool, watching five thousand early adopters discover that permissionless finance could reach people the banks had spat out — and watching the same machinery attract wash traders and predatory bots within weeks. I ended up retreating to a cabin in the Alps for two weeks, not to celebrate, just to breathe, because the gap between the ideal of financial freedom and the reality of speculative extraction had become physically painful to hold. What I carried back is what I'd offer to anyone holding a token through this winter: the protocol's philosophy and the protocol's behavior are two different assets, and they rarely move together. Judge the behavior.
Now the third thread — and this is where the résumé stops being about either crypto or AI and becomes about us.
In 2021 I spent months tracing the on-chain metadata of a generative art project that had sold the dream of permanent, decentralized ownership. The metadata lived on centralized servers. The permanence was a performance. My exposé went viral in privacy circles, and a good number of people accused me of killing the culture, which taught me that truth tends to isolate you before it liberates anyone. Three years later I started working — with a protocol called SynthVoice — on the opposite problem: proving that a human is actually present in a stream of content that might be entirely synthetic. I wrote a manifesto called "The Proof of Soul," arguing that in an age of manufacturing, cryptographic identity is the last honest anchor of the human. We secured half a million in grants. But the point wasn't the money. The point was that I had finally found the through-line connecting the audit, the lending protocol, the metadata exposé, and the coming flood of synthetic media: every one of them is a question about who or what you can trust when the surface looks perfect.
Hubinger's alignment work is downstream of that exact anxiety. So is Interledger. So is the entire trust-minimization project. The crypto people who left for AI didn't abandon the mission. They followed it to where the stakes had grown larger than a balance sheet.
Here is the contrarian turn, and I'll make it uncomfortable. The industry's favorite reading of this story is that crypto lost its talent, and the joke is a verdict. I think that's backwards. The joke is a defense mechanism.
The XRP community didn't laugh because the story is funny. It laughed because self-deprecation is the cheapest way to keep a room warm when there's no new narrative to heat it. When you can't rally, you roast. It feels like a community. It functions like anesthesia. And the specific thing being numbed is the question of whether the protocol you hold still has the people who could meaningfully improve it. Note how quickly the conversation slid from "the alignment lead once worked here" to "but XRP hasn't pumped" — a pivot that converts a talent signal into a price complaint, because price complaints are survivable and talent droughts are not.
The second contrarian point cuts against my own tribe. The crypto-to-AI migration is not evidence that crypto failed. It is evidence that crypto succeeded at producing a specific kind of mind — one trained to reason about adversarial trust, to assume the counterparty is hostile, to demand that guarantees be verifiable rather than promised. That mind is now in enormous demand, because AI safety is the biggest adversarial trust problem ever posed. We should be proud that we built it, and honest that we couldn't hold it. Pride without honesty is just another form of the anesthesia.
So watch what the trustless do next, not what the chart does. The people who spent a decade removing the need for intermediaries are now deciding what to do about an intelligence that refuses to be inspected. If cryptographic identity becomes the proof that a human still stands behind the words — the Proof of Soul made practical — then the résumé everyone laughed at will read less like an exit and more like an early map. The question isn't whether crypto can afford to lose its alignment engineers. It's whether it can afford to keep ignoring why they left.
