Can AI break crypto wallet security?
The full reading behind my Short: what was said, who said it, what's proven and what isn't, and calm steps for ordinary holders. Not financial advice.
The short version
On 6 October 2026 OpenAI released 722 maths manuscripts produced by an unreleased AI model [3]. On the All-In podcast a few days later, David Friedberg pointed out that there were no major cryptography results among them and speculated that some may have been held back. He warned that if AI cracks the maths behind wallet keys before quantum computers do, "then we have a real problem" [1]. Separately, Ethereum researcher Justin Drake urged the industry to "calmly start planning for 'bunker mode'" [3].
What we know: no practical attack on Bitcoin or Ethereum wallet keys has been demonstrated [3]. This is a warning, not a proven attack. A more proven risk is AI getting better at finding bugs in crypto software [3][4][5].
How wallet "locks" work
A crypto wallet uses a secret number, the private key, to approve payments. A related public key lets the network check those approvals. Getting the public key from the private key is easy; working backwards is supposed to need an impractical amount of computing [3].
Bitcoin and Ethereum signatures use maths called elliptic curves. The worry is that AI might find a shortcut, so someone could work backwards on ordinary computers, without the quantum computers the industry has been preparing for [3].
Watch the Short
About this page
The Short uses my own animation only: no podcast footage, audio, logos or likenesses. Quotes are short and credited. I don't recommend any coin, and this is not financial advice.
What was said, and by whom
OpenAI's maths release
OpenAI released 722 manuscripts from an unreleased model tested on about 4,000 research problems. OpenAI said some results have computer-checkable proofs, while others are unverified and could contain errors [3].
David Friedberg (All-In podcast)
In the episode released 10 Oct 2026 (around 50:09–51:36), Friedberg said people had noticed no major cryptography proofs in the release and suggested this might mean there were breakthroughs that weren't published. He described "a bit of a panic underway in the crypto community", said "the rumor is" that OpenAI has more releases coming, and added: "if it is a risk, it needs to be made known" [1][2].
A sceptical view on the same show
Chamath Palihapitiya called the proofs impressive but described them as "very small, very narrow, specific bodies of intellectual exploration", and doubted they would quickly lead to big real-world breakthroughs [1].
Justin Drake (Ethereum researcher)
Drake wrote: "Today I call upon the blockchain industry to calmly start planning for 'bunker mode'." In his worst case, AI could break wallet signatures "in months, not years". He suggested large holders gradually move funds to addresses whose public keys have never appeared on-chain [3].
Vitalik Buterin (Ethereum co-founder)
Buterin agreed the risk is real, but told holders not to rush, warning that mistakes during a migration can cause losses of their own: "I personally have lost more money in botched migrations than I have lost in all hacks combined" [3].
What's proven, and what isn't
- Not proven: a break of wallet keys. No practical attack on Bitcoin or Ethereum wallet keys has been demonstrated, and none appeared in the research CoinDesk reviewed [3].
- Not proven: hidden cryptography results. The idea that OpenAI is holding back cryptography results is speculation. Friedberg himself framed it as what "the rumor is" [1].
- Worst case, not a forecast. Drake's "months, not years" is his worst case. The Ethereum Foundation's target for moving to quantum-resistant cryptography is December 2029 [3].
- Better evidence: AI finding software bugs. In OpenAI's EVMbench test, the best AI agent detected 45.6% of 120 known smart-contract vulnerabilities, and another exploited 72.2% of a curated subset in a test environment [4]. A follow-up study found results unstable: on 22 real incidents newer than the models, agents detected up to 65% of vulnerabilities but none achieved a full end-to-end exploit (0 of 110 attempts) [5].
- Real losses linked to bugs. CoinDesk reports recent cases where crypto software flaws were exploited and companies said they suspected AI helped find the flaw [3]. That is a software risk, not a break of the underlying maths.
Calm steps for ordinary holders
- Don't panic or rush. Rushed wallet migrations can lose money on their own (Buterin) [3].
- Follow official updates from your own wallet provider or exchange, through their official app or website.
- Ignore urgent messages. No legitimate business will demand you send crypto "to protect your money"; that's always a scam [6].
- Don't click links in unexpected messages, even if they seem to come from a company you know [6].
- Large or experienced holders: Drake's suggestion is a gradual move to fresh addresses whose public keys have never appeared on-chain [3]. If you're unsure how, wait for guidance from your wallet provider rather than improvising.
This is general information, not financial advice. No price predictions, no coin recommendations.
Full narration (transcript)
Could AI crack the locks on crypto wallets? OpenAI has just released over 700 maths papers made with its AI. On the All-In podcast, David Friedberg noted no big cryptography results, and wondered if some were held back. He said if AI cracks wallet maths before quantum computers do, then we have a real problem. Ethereum researcher Justin Drake has urged the industry to calmly plan for a bunker mode. But no one has shown a working attack. This is a warning, not a proven attack. So don't panic or rush. Follow official updates from your wallet or exchange. Not financial advice.
Sources
- [1] All-In Podcast (10 Oct 2026). "Is Claude Conscious? Pope Rejects, Model Welfare Movement, OpenAI's Math Backlash, France Riots" (about 50:09–51:36). YouTube · transcript (Podscripts)
- [2] Bitcoin News on X (10 Oct 2026), attributing the cryptography comments to David Friedberg. Post
- [3] Malwa S. (8 Oct 2026). "Bitcoin and ether holders urged to prepare 'bunker mode' against possible AI attacks." CoinDesk. Article
- [4] Wang J. et al. (2026). EVMbench: Evaluating AI Agents on Smart Contract Security. OpenAI, Paradigm, OtterSec. PDF
- [5] Peng C., Wu L., Zhou Y. (2026). Re-Evaluating EVMBench: Are AI Agents Ready for Smart Contract Security? arXiv:2603.10795. arXiv
- [6] US Federal Trade Commission. What To Know About Cryptocurrency and Scams. consumer.ftc.gov
Credits
Video and images: own animation (padlock, chip and charts are illustrations). Quotes are short and credited to their speakers.