How OpenAI's Math Dump Has Crypto Rethinking Its Cryptography: Uneasy Money

Ken Ward said OpenAI released 722 papers, some formalized in Lean, renewing concern that AI could help find ways to break cryptographic systems. The work did not show that elliptic curves had been solved, and the speakers characterized the threat as unprecedented but uncertain. They described model progress as rapid and uneven: systems that struggled with simple arithmetic two years ago can now do substantial mathematical work, though their abilities remain “spiky.” Some model-generated results may rely on computational or brute-force exploration. As far as one speaker knew, expert mathematicians can still review and understand the output; the next six months may offer more evidence about whether that changes, but past predictions about AI capabilities have often been wrong.

The immediate question is whether growing compute could eventually make elliptic-curve keys vulnerable to brute force, not whether they are broken now. If keys could be cracked within a reasonable human timeframe, one speaker said, elliptic-curve cryptography would effectively be broken, though even a severe scenario would require substantial time and computing power. Justin Drake’s warning led to suggestions such as moving assets above 50 BTC to wallets protected by a different hash function. The speakers cautioned that rushed cold-wallet rotations could cause holders to lose funds before a GPU farm could steal them. Large balances may be earlier targets because attacks are costly, especially assets held by infrastructure providers in concentrated addresses. A cited $5.5 billion WET address is a smart contract, however, not a conventional private-key-controlled wallet.

The potential exposure goes beyond crypto: elliptic-curve cryptography also supports HTTPS and internet traffic. If the threat is real, speakers thought powerful governments or other geopolitical actors might be likelier early users than thieves pursuing Satoshi’s coins, and they assigned relatively low odds to the rumors being true. Crypto networks could be targeted first, but may be able to patch node software, fix bugs, or roll back changes; legacy financial institutions may respond more slowly. Open-source code is visible to attackers, but broad scrutiny and participation could make open systems more secure over time. Reports that OpenAI is coordinating with the Ethereum Foundation were described as unconfirmed.

A separate MetaMask staking incident involved 0.36 ETH in fees or rewards being redirected from previously specified addresses to new ones. Speakers distinguished staking withdrawal keys from validator keys and said fee destinations are a separate server-side setting. One speaker judged validator-fund theft risk low because withdrawal keys are not kept on servers. MetaMask had not yet given a full explanation, and a postmortem was expected after the incident was resolved.

Zach befriended Jimmy Green, whom the speaker identified as a Chinese money launderer for the DPRK, and conducted undercover work to obtain addresses and trace laundering networks. The speaker said DPRK-linked operators laundered more than $1 billion in four or five months; Jimmy also claimed Kim Jong-un had released 50 Bitcoin that day. The operators reportedly knew they were working with the DPRK and used crypto-industry language about decentralization and Bitcoin maximalism to advance their operations. Relevant public- and private-sector contacts had been informed before publication, and the speakers argued that reporting could spur action and help exchanges recognize laundering patterns. Chinese-language account indicators alone do not rule out DPRK coordination. They put Zach’s investigation spending at either $350,000 or about $400,000, giving conflicting figures.

The Gnosis and Safe discussion centered on the value created by long-running projects, despite failed products and experiments. One speaker recalled Gnosis raising 330,000 ETH in its 2017-era ICO and estimated it still held about 180,000 ETH, while acknowledging that the recalled valuation and token-sale figures had not been checked. Safe, which followed the Parity multisig hack, was described as securing more than $100 billion. Reported funds in Safe accounts fell from $66 billion to $30 billion, for reasons the speakers did not know; asset-price changes were one possible explanation. Safe’s broad adoption was attributed partly to charging no fees, while off-chain infrastructure complicates monetization and a fee could invite a no-fee fork.

As regulatory fears recede and speculative premiums fade, speakers said token projects will face greater pressure to return value to holders through mechanisms such as fee distributions or programmatic buybacks. They argued that governance-token value can also contribute to economic security: if a token is worth less than the cost of stealing protected assets, a protocol may be exposed. Charging for wallet convenience or enterprise security could be sustainable but risks alienating users who expect basic security to be free. The speakers also argued that DAO disputes and lawsuits do not by themselves prove governance has failed; governance is meant to resolve conflict, resist capture, and create accountability. Separately, they viewed the shutdowns of Blast and Abstract as possible end-of-cycle capitulation, while saying L2s need to differentiate beyond offering a general-purpose EVM.