CoinGlass lists it plainly: no circulating supply, no spot market, no float. Yet the Anthropic pre-IPO perpetual on Binance carries roughly $79 million in open interest, with 24-hour volume north of $20 million. The chart doesn't lie about one thing โ capital is moving. It lies about almost everything else.
I have spent a decade auditing contract logic and mapping liquidation cascades. When I pulled the specification for this instrument, I found something rare even by 2026 standards: a derivative with no underlying asset, no on-chain settlement, and no convergence mechanism. This is not a blockchain product wearing financial engineering. It is a financial product wearing blockchain aesthetics โ a cash-settled perpetual swap on a company that has never issued a tradable share.
That distinction matters more than the headline number. Here is the forensic chain.
Anthropic filed a draft registration statement with the SEC in June โ confidentially. It is still private. No public float, no ticker, no shareholder register a retail trader can reach. Into that vacuum, centralized exchanges have built synthetic perpetuals letting traders bet on what Anthropic might eventually be worth.
The mechanics look standard. A perpetual contract uses a funding rate to periodically transfer value between longs and shorts, nudging the contract price toward a reference. On Bitcoin, that reference is a 24/7 spot market where arbitrageurs hedge in seconds. On the S&P e-mini, it is a cash-settled index with a defined settlement window and institutional desks enforcing the basis. Both work because a physical anchor exists.
Here, the reference is a valuation nobody can verify until an underwriter prices a book. Binance Research โ the exchange's own arm, and therefore a conflicted source โ describes the contract as needing no underlying shares. In their own framing, traders are taking opposing bets on a company's final value. Open interest across Anthropic and OpenAI contracts now exceeds $160 million. Binance accounts for roughly 40% of activity, which tells you liquidity is concentrated in a single venue with a documented history of regulatory settlements. Anthropic itself has said nothing. It did not endorse this market, does not provide data to it, and holds no obligation to it.
That is the first red flag. When I audited ERC-20 implementations in 2017, my rule was simple: if the counterparty won't sign the specification, you are not trading their asset. You are trading a story they happen to be the subject of.
Here is the part launch coverage smooths over. Every perpetual is a zero-sum instrument wrapped in a fee layer. For every long there is a matching short. Open interest measures total notional still open, not net conviction. When a research note cites record open interest as evidence of bullish sentiment, it commits a category error. Rising open interest means rising participation and rising leverage. It says nothing about direction.
Forget the narrative. Follow the mechanism.
A perpetual with no spot leg cannot converge. The funding rate is the only tool binding a perpetual to reality. It pays the crowded side and charges the other, creating an incentive for arbitrageurs to take the opposite position and pull price back. That arbitrage requires a spot leg to hedge. Remove the spot leg and the funding rate becomes a sentiment tax, not a price signal. Longs can pay shorts indefinitely while the contract drifts. Nothing forces convergence, because nothing can be bought or delivered. The mechanism is not weakened here. It is absent.
There is no settlement event to anchor terminal value. A futures contract converges at expiry because someone must deliver or cash-settle against a defined index. A perpetual has no expiry by design โ it substitutes the funding mechanism for that anchor. Strip both, and fair value is entirely reflexive: the contract is worth what the next trader will pay. I mapped this exact failure mode in 2022 while doing forensics on 850,000 Terra wallets. The redemption mechanism looked like a peg. It was a queue. When the queue broke, solvency failed at a precise block height and $40 billion evaporated. A stability mechanism that depends on continuous new participation is not a mechanism. It is a countdown.
The regulatory classification is not ambiguous. Apply the Howey test: capital contribution (yes), common enterprise (yes โ shared exposure to Anthropic's value), expectation of profit (explicit), and reliance on others' efforts (the company's management and IPO timing). Four for four. That is a security swap in everything but the label. The venue offering it sits outside the registration regime governing the underlying company. This is textbook regulatory arbitrage โ a wrapper engineered to sit in the gap between a private company and public securities law.
The exchange's own research draws a comparison to tokenized equities. Dismantle it. Tokenized stocks carry actual securities in custody; the token is a claim on a real asset, and the issuer holds the underlying. A pre-IPO perpetual holds nothing. No custodian, no share, no legal claim on Anthropic equity. One is a receipt for an asset. The other is a receipt for an opinion.
The size is trivial, and that is the point. Anthropic's reported valuation ambition is north of $2 trillion. Against that, $80 million in open interest is 0.004% โ a rounding error. Even the combined Anthropic-plus-OpenAI figure is four orders of magnitude below the equity valuation it supposedly "discovers." A market with no spot leg, no settlement, and 0.004% of the capital base is not discovering price. It is discovering sentiment about price.
Now the data issue, which I treat as a finding in its own right. One circulating dataset claims open interest grew from roughly $1 million in April to more than $160 million now, described as "179% growth." Run the arithmetic: $1 million to $160 million is 160x, or +15,900%. The stated 179% cannot describe that series. Either the growth figure is mislabeled, or it refers to a different base period, or one number is simply wrong. A 15,900% error does not survive a single sanity check. The source is conflicted and the dataset is internally inconsistent โ a two-layer credibility failure, buried under a hype cycle.
On governance, there is none. The exchange sets the rules, the oracle, the margin schedule, and the delisting policy unilaterally. Participants have no vote and no recourse. I have spent years documenting on-chain governance turnout that rarely clears 5%, with a handful of whale wallets deciding outcomes regardless. This product is worse โ it does not even pretend. No proposal process, no parameter vote, no forum. The counterparty can reprice your position by editing a configuration file. That is not decentralization with low participation. That is a private bookmaker with a terminal.
And there is nothing to audit on-chain, which is itself the warning. On-chain data doesn't lie โ which is precisely why this product isn't on-chain. Smart contracts have no mercy because they cannot lie: the code executes identically for every address, and the ledger remembers everything. This instrument deliberately opts out of that property. Pricing rules, liquidation thresholds, and the reference feed are undisclosed. When I audited 45,000 lines for a 2017 ERC-20 launch, I insisted on a standardized regression suite before mainnet. The only reason to refuse that transparency is that the transparency would be unflattering.
Here is what a real analyst would track, and what is conspicuously missing. In a perpetual without a spot leg, the funding rate is the single observable revealing positioning. Extreme positive funding means longs are paying to stay long โ a crowded, fragile book. Extreme negative means the opposite. Neither number is consistently disclosed on this contract. An instrument whose only internal signal is unpublished is not a market. It is a black box with a price tag.
I built a framework in 2026 to classify 200,000 AI-agent transactions across L2 networks, measuring gas cost against success rate. The finding that stayed with me: 12% of network congestion came from poorly optimized automated scripts grinding through loops no human designed. Momentum strategies pile into a narrative, the funding rate spirals, and retail followers get filled at the worst tick. The congestion here isn't block space. It's order flow.
The prevailing narrative calls this "price discovery ahead of the IPO." I want to push back, because the corollary doesn't hold. Crypto traders, the story goes, are pricing Anthropic before Wall Street can. But price discovery requires a price that aggregates information about the underlying. Here the underlying has no market to inform it, the contract has no spot leg to hedge against, and participants are speculating on media sentiment rather than revenue, margins, or model performance. What this market publishes is not a forecast of Anthropic's value. It is a forecast of what crypto traders think other crypto traders think about Anthropic. That is a sentiment loop three layers deep, and loops amplify noise rather than discover truth.
Correlation is not causation, and here it may not even be correlation. The contract's moves track product announcements and IPO-timetable headlines โ OpenAI tools rising on an Astra model release, falling on hints of a delayed offering. Those are narrative inputs, not financial inputs. A market that only moves on headlines has ceded pricing to the news cycle. The honest framing: this is a shadow casino wearing a valuation's name. That is not a moral judgment. It is a mechanical description.
Watch two numbers next quarter, and neither is the price. First, the funding rate โ if it goes undisclosed or stays extreme, the book is structurally unstable. Second, the SEC file trail โ Anthropic's offering has reportedly slipped from October toward November, and every further delay removes value from the narrative propping this contract up.
The ledger remembers everything. This instrument isn't on a ledger. That absence is the entire story.