
Researchers report an 86% reduction in the cost of cracking Bitcoin's elliptic curve signatures using AI-driven coding agents.
AI-generated summary
Bitcoin and Ethereum use the secp256k1 elliptic curve for security, which is theoretically vulnerable to quantum computing. Researchers are using AI agents to measure the resource requirements for potential attacks to better inform defensive strategies.
Morning Minute is a daily newsletter written by Tyler Warner. The analysis and opinions expressed are his own and do not necessarily reflect those of Decrypt. And check out our daily news show 'FOMO HOUR' covering all of the top stories and market action.
GM!
Today’s top news:
Crypto majors fall another 1-3% ahead of CPI; BTC at $77k
BTC ETFs see $280M in outflows and $440M on the week
Binance’s Bitcoin reserves hit 2-year highs
Stonkfun ecosystem soars as Stonk hits $0.30 and multiple pairs run 2-4x+
Robinhood Chain leaders rebound as volumes level off; Flycoin soars to $50M in its debut
🤖 AI Agents Cut the Cost of Cracking Bitcoin’s Cryptography by 86% in Two Months
More than 100 researchers working alongside AI coding agents significantly decreased a benchmark for a key step in a quantum attack on Bitcoin and Ethereum from 10.75 billion down to 1.496 billion. That happened between late May and July 26, but the paper landed just this week.
For background, Bitcoin and Ethereum both secure signatures with secp256k1, an elliptic curve. A sufficiently powerful quantum computer could reverse that math and derive a private key from a public one. That’s the problem. The competition, called ECDSA.Fail and run by Eigen Labs, scored circuit designs by multiplying logical qubits against Toffoli gates, an expensive quantum operation. A lower score means fewer resources needed. The leading design used 1,151 logical qubits and roughly 1.3 million Toffoli gates, and a later submission pushed gates below a million. That leading score came in at about half of Google Quantum AI’s March benchmark, though the two used different counting methods and don’t compare cleanly.
The paper was authored by researchers from Eigen Labs, Trail of Bits, StarkWare, Theta Labs, MultiVM Labs, and the Ethereum Foundation. That’s crypto’s own security establishment funding and publishing work that makes the attack cheaper, on the theory that knowing the real cost is the only way to plan the defense. They call the method Open Autoresearch, humans and AI agents iterating against a shared measurable target with a verifier in the loop.
This lands in the middle of a quantum scramble that’s been building for weeks. The Ethereum Foundation set a hard December 2029 deadline to make transactions, validators, and storage quantum-resistant. StarkWare pushed the first quantum-safe Bitcoin transaction to mainnet. Ethereum developers proposed rebuilding the validator deposit contract. Ripple is hardening the XRP Ledger. Galaxy committed up to $5 million in July, and nine firms including BlackRock, Coinbase, and Strategy pledged $15 million over three years. The researchers point to NIST’s draft proposing deprecation of classical public-key algorithms at the 112-bit level after 2030 and disallowing them after 2035.
So where does this leave us? The defense side has a fixed schedule measured in years. The attack side just got 86% cheaper in two months because a hundred people pointed AI agents at it. That’s not a prediction about Q-Day, and no funds are at risk today. But every timeline in this space was built assuming attack research moves at human speed, and this is the first hard evidence it might not.
🌎 Macro Crypto and Markets
Crypto majors are red down 1-4%; BTC -1% at $77k; ETH even at $2,455; SOL -2% at $100; HYPE -4% at $79; ZEC -10% at $1,095
Top alt movers include RAY (+13%), PONS (+11%) and FF (+5%).
Oil +2% at $99; Gold -1% at $4,385
Stock futures are green after oil sold off from highs; DOW +0.6%, Nasdaq +0.6%
Senate Republicans released a revised 630-page Clarity Act ahead of a September 15 procedural vote, adding CFTC registration for “decentralized-in-name-only” protocols while leaving ethics provisions largely unchanged
Sam Bankman-Fried petitioned the Supreme Court to overturn his fraud conviction, arguing he was barred from showing customers lost nothing and calling the $11 billion forfeiture an excessive fine
Brian Armstrong told CNBC that $400,000 Bitcoin by 2030 remains a reasonable target, saying the year-long down cycle has bottomed and expecting SEC and CFTC guidance within weeks
Binance's Bitcoin reserves topped 693,000 BTC, a two-year high and roughly 30% of major exchange reserves, after about 77,000 BTC moved onto the platform since late April
Singapore's Alpha Ladder began distributing Payward's xStocks to institutional and accredited investors across selected APAC markets through its WealthX platform, following an MOU with Payward and MetaComp
Ethereum developers tentatively set the launch date for Glamsterdam for Sepolia on October 6, though the upgrade hasn't run stably on any private devnet and no Hoodi or mainnet dates are fixed
Corporate Treasuries & ETFs
The Bitcoin ETFs saw $283M in net outflows on Thursday; the ETH ETFs saw $30M in outflows
Meme Coin Tracker
Meme leaders were red; DOGE -2%, SHIB -2%, PEPE -4%, PENGU -3%, TRUMP -2%, SPX -2%
Robinhood chain leaders rebounded on the day; Pons +10% to $420M; AI +25% at $250M; Cashcat even at $160M; Flybrain +830%, Zforge 100x, Juggernaut +600% and Frong +100% led top movers
On BSC, Niu Lau jumped 70% to $130M after its Binance spot listing; BREW +100% and Marscoin +20% were other top movers
Solana top movers were led by Ember +165% and Fone +40%; Ansem -7% at $150M
The Stonkfun ecosystem soared last night as STONK broke $0.30; Stonk +55% at $270M, ZCAT -7% at $83M, KNOTS +100% at $44M, BTC +94% at $18M and PURR +43% at $13M; LOOP +100x, METH +7x, LMEOW +180% , Stonkcat +7x and JUPCAT +6x led top movers
💰 Token, Airdrop & Protocol Tracker
Hyperliquid led the onchain protocols with $2.08M in revenue on the day; PUMP was 7th with $1.4M, PONS next with $1.14M and Stonk was 13th with $760k
Onchain RWA perpetual volume passed $120 billion in August, up from under $1 billion in October 2025, with open interest near $4.9 billion
Raydium’s RAY token is up 86% this week on the back of its connection with the growing Stonkfun ecosystem
🚚 What is happening in NFTs?
AI outlook — possibilities, not facts
Ethereum developers to continue work on quantum-resistant validator upgrades.
Likely · Within months

Anthropic reported that Russian- and Chinese-speaking operators, as well as an independent consultant linked to Mali's intelligence, used its Claude AI model to automate cyberattacks, vulnerability research, and a population-scale surveillance system.

Blockstream refused a ransom demand from hackers holding about 598 Bitcoin on the Liquid Network, labeling the theft a crime rather than responsible disclosure and vowing to work with law enforcement.

Hardware-wallet makers Trezor and BitBox warned users on Sept. 9 about phishing emails impersonating their brands following third-party email provider breaches. Both companies urged recipients to avoid suspicious links and keep recovery seeds private.

XRP Healthcare is winding down operations after a flaw in its XRPH Wallet exposed thousands of accounts and led to roughly $450,000 in stolen assets. The breach stemmed from improper entropy input and use of Math.random(), drastically reducing the keyspace. The company will delist tokens, keep IP, and continue recovery efforts despite shutdown.

OpenAI reported that 10,000 concurrent AI agents solved a Navier-Stokes fluid-motion problem in 88 hours, with formal verification in Lean taking another 17 hours using GPT-6 Astra. The breakthrough establishes cases C and D of the Millennium Prize formulation and could automate theorem proving for smart-contract security, reducing labor-intensive verification while increasing the importance of accurate specification design for DeFi protocols and tokenized assets.

OpenDesign Arena benchmarked 13 AI models on real-world design tasks, with GPT-6 Astra scoring highest at 82.7 points and DeepSeek V4.1 Flash close behind at 81.2 points while costing only $0.023 per design versus $1.61 for the leader. DeepSeek's model uses a Causal Encoder-Decoder architecture activating only 8-16 billion of its 552 billion parameters, enabling low cost and fast delivery.