The headline arrived without a byline, without a dateline, and without a single quotation mark. It said that the presidents of the United States and China had agreed to rename artificial intelligence "superintelligence," and that the renaming would boost infrastructure. It ran on a crypto publication. It contained no crypto.
I read it twice โ not because it persuaded me, but because of how efficiently it worked. Four high-arousal tokens arranged into one sentence, doing labour that once required a leaked memo, a corroborating source, and a wire service. No name, no date, no spokesperson, no joint statement; just a noun traded up for a larger noun, and two heads of state positioned behind the trade.
To hunt the truth, one must first bury the hype. So let us bury this one properly, because what it leaves exposed is worth more than the story ever was.
Context: The Information Economy of a Bear Market
Bear markets are information-poor and narrative-rich. When capital is scarce, the cost of manufacturing a catalyst falls below the cost of verifying one โ and 2026 has pushed that asymmetry to an extreme that should bother anyone carrying risk. The distribution layer is now a language model plus a publishing template. The verification layer is still a person with a browser.
That mismatch โ a crypto venue publishing a geopolitical AI claim with no crypto in it โ is the first diagnostic. It points to traffic arbitrage rather than reporting: take a term the audience already reacts to, attach it to a political figure the audience already reacts to, and let referral traffic settle the question of whether the claim deserved checking. The article carried no author. That is not an oversight; anonymity is a governance choice. It shrinks the surface area for correction and removes a person who could be asked a question.
I have watched this curve bend across four cycles. In 2017, drawing on my audit work in Barcelona, I read more than fifty ICO whitepapers and found the same flaw in almost every one: a token whose utility was asserted entirely by its name. I called it the utility-token fallacy, and I was early and unpopular for it. In 2020 I dug into Uniswap's liquidity incentives and concluded the real architecture was a social contract, not a bonding curve โ trust encoded in behaviour rather than in code. In 2021 I argued that NFTs would matter as credentials, not collectibles. In 2022 I wrote about the cost of belief and meant it literally. By 2025 I was translating regulatory language for institutions under the banner of compliant decentralisation.
Each cycle, the fabrication cost of a catalyst fell. In 2017 a fake catalyst cost a whitepaper and a Telegram channel. In 2020 it cost a yield number with no cohort analysis behind it. In 2026 it costs a prompt.
Core: What the Story Is Actually Made Of
Strip the framing and five claims remain. Two heads of state reached an agreement. The agreement was to rename AI. The new name is "superintelligence." The purpose was to boost infrastructure. The effect will be more political support and more investment. Not one of those five claims carries a verifiable anchor โ no readout, no spokesperson, no text, no budget figure, no date. A head-of-state-level understanding is not an anonymous thing. It always arrives with a name attached, because someone always wants credit.
Consider what a genuine bilateral deliverable looks like, because it is dull in a very specific way. Readouts are written in the passive voice of diplomacy: the two sides exchanged views, took note, agreed to maintain communication. Real agreements have legal texture โ a memorandum of understanding, a working group with a reporting cadence, a visa category, a mutual-recognition clause, language on export controls.
The verifiable ChinaโUS AI story is not cooperation but friction: compute export controls, advanced-node restrictions, investment screening. A narrative in which the two powers jointly rebrand a field to "boost infrastructure" does not merely lack evidence; it contradicts the structural reality it claims to describe.
And ask what a renaming could possibly implement. International agreements operate through instruments that carry operative force โ tariff schedules, export control lists, entity designations, screening thresholds, visa allocations, standard-setting bodies. A name carries none. Renaming a technology does not alter one export licence, does not reclassify one chip, does not move one data centre forward in an interconnection queue. When a claim routes around every instrument that could implement it, you are not reading a policy story. You are reading a preference.
What would make the claim analysable is precisely what is missing. Which infrastructure โ compute clusters, grid capacity, optical interconnect, chip fabrication? Which instrument โ a joint statement, a memorandum, a licensing change? Which date, which forum, which official? The absence is not a gap in reporting. It is the design.
Then there is the technical vacancy. The article contains no architecture, no training method, no dataset, no compute specification โ nothing that could be analysed even generously. "Superintelligence" appears as an object to be named rather than an entity to be defined, with no capability boundary drawn. The piece also collapses AGI, ASI, and large language models into a single unmarked category. In serious usage, superintelligence denotes a hypothetical system exceeding aggregate human cognitive performance, in the tradition Bostrom set out. Using it to describe a policy conversation does not elevate the conversation; it spends the term's precision to buy a headline.
The vocabulary is the tell. "Superintelligence" was almost certainly harvested, not sourced โ pulled from the 2024โ2025 public discourse of frontier labs that adopted the word as a strategic label. And "infrastructure" is a financialised term, not an engineering one. Engineers specify megawatts, GPU counts, interconnect queues, capital expenditure schedules. Bankers say infrastructure, because it rings the widest possible set of tickers without committing to any of them. That same word choice saturates the RWA pitch I have been reading for three years โ a storytelling exercise whose underlying truth is that most traditional institutions have no operational need for a public chain, and never did.
The pattern generalises, which is what should worry you. A data availability layer that no rollup generates enough data to saturate. A post-halving mining sector whose revenue curve no longer funds the hashrate it built, and whose hash power keeps drifting toward a shrinking set of pools. A token whose entire thesis is its ticker. In each case, the name substitutes for the mechanism, and the market prices the name.
Now trace the distribution. The piece runs on a crypto outlet, a venue whose audience is primed for asymmetric catalysts and whose editorial incentives reward volume over verification. From there, cross-border re-upload and translation strip the hedges first, then the qualifiers, then the attribution. Within a week, "reportedly discussed AI cooperation" becomes "agreed to rename AI." The chain of custody does not merely lose detail; it manufactures conviction. Information that arrives pre-digested and pre-believed is worth more to a distributor than information that arrives checkable.
And the market rewards it. Narrative-to-ticker latency is measured in tens of minutes now, and in a thin order book small flow moves price. When I audited DeFi incentive structures, I learned to ask who pays for the liquidity โ and the answer was never the people holding the bag at the end. The equivalent question here is simpler. Who paid for this headline, and who was holding when it printed?
Behavioural economics has a name for the mechanism: an availability cascade, in which a claim becomes credible through repetition rather than evidence, and each repetition lowers the cost of the next. Layer on the friction asymmetry โ checking takes a browser, a source, and a willingness to look foolish for doubting; sharing takes a tap โ and the outcome is close to deterministic. The rational move in the moment is to pass the claim along. The rational move across a cycle is to be the person who did not.
Contrarian: The Burying Is the Job
Here is the uncomfortable part. The story being false is not the news. The news is that it never needed to be true to do work. A fabricated readout that moves nothing is a curiosity. A fabricated readout that moves anything proves that the asset being traded is not the protocol but the sentence someone wrote about it. That distinction is the entire bear market.
The outlet, meanwhile, is not an outlier. It sits at the far end of a continuum whose respectable end is unlabelled sponsored content, "per reports" aggregation, and analysts summarising each other's summaries. The distance between the edge and the middle is shorter than the industry admits, and it is measured in degrees of hedging rather than in kind. Fact-checking cannot scale against syndication; the check takes hours, the propagation takes minutes. What scales is provenance โ signed sources, verifiable identifiers, credentials that travel with a claim instead of being stripped by the translation layer. That an industry which spent a decade building tamper-evident ledgers has not built this is not an irony. It is a choice of incentives.
And had the story been true, you still could not have traded it. "Boost infrastructure" does not map to a single instrument, supply chain node, capital figure, or timeline. A headline that names everything is usually telling you it knows nothing. That is the signature of content built for breadth rather than accuracy โ the same signature I found in the 2017 pitch decks, where a token could do anything and therefore meant nothing.
To hunt the truth, one must first bury the hype. In this market, the burying is the job; the hunt is what you do with the hours you saved.
Takeaway
Over the next seventy-two hours, exactly one thing settles this: whether any official channel โ the White House, the Chinese foreign ministry, the commerce ministry โ publishes anything at all about AI. If nothing appears, the correct conclusion is not that the story was exaggerated. It is that it was manufactured, and that the manufacturing worked, at least once, on someone.
Which raises the question worth carrying into the next cycle. If a headline with no author, no date, and no source can still move a ticker, what exactly do you own โ the protocol, or the story someone else wrote about it?