A gold bitcoin coin standing on a dark surface, with a blurred reflection of the same coin behind it, as if a signature had been repeated
Our own AI-generated illustration. It does not represent real data.

Reuse an ECDSA nonce and the private key is no longer secret: 3,278 bitcoin keys, and bots emptied bait in a second

Researchers at the University of Liège recovered 3,620 private keys because wallets reused the nonce in ECDSA signatures. 3,278 of those keys were on bitcoin. The addresses' peak historical balances add up to €101 million, but that is not a confirmed theft. When the team planted a few dollars of bait, bitcoin bots arrived within a second.

At 8:30 a.m. Eastern on November 20, 2026 (2:30 p.m. in Madrid), CoinDesk reported the work of Vincent Jacquot and Benoît Donnet, of the Montefiore Institute at the University of Liège [1][2]. Their paper, "Oops… I Did It Again. I Reused my Nonce," was presented at Financial Cryptography and Data Security and is posted in the university repository [2]. The authors scanned ECDSA signatures on six chains, from each network's genesis through June 30, 2024 [2].

The finding is not that ECDSA itself has been broken. It is that an old implementation mistake still shows up, and that on bitcoin, automated programs are waiting for it.

What a nonce is, and why it cannot be reused

That is a different problem from having the public key in the open, the issue we covered in the bunker-mode piece. There the private key is still secret, even if the public key is no longer hidden behind a hash. Here, once the nonce is reused, the private key is no longer secret at all.

What they found on six chains

Run separately on each network, the attack recovered 3,605 distinct keys. Three of those keys were used on more than one chain, so the unique count is 3,605 rather than 3,608 [2]. Combining signatures from all six networks turned up 15 further keys, for a total of 3,620, because the same nonce had been reused across blockchains [2]. At their peak, those extra 15 keys held a little under €2,000 [2].

Chain Signatures with a reused nonce Private keys recovered
Bitcoin 2.50 million 3,278
Dogecoin 578 195
Ethereum 21,300 119
Litecoin 26 9
Bitcoin Cash 141 6
Dash 11 1
Total (within each chain) 2.90 million 3,605

Source: Table 1 of the paper, the within-reuse count through June 30, 2024 [2]. The headline total of 3,620 includes the 15 keys that appear only when the six networks are searched together.

All amounts in the study use prices as of November 25, 2024 [2]. Adding up each vulnerable address's peak balance gives €101 million, about $106 million, the authors told CoinDesk [1][2]. The sum assumes an attacker reached every address at its richest moment and does not sort out later deposits. Jacquot told CoinDesk the team did not establish how much was actually stolen [1]. It is historical exposure, not a confirmed haul.

Most of that exposure dates to bitcoin's early years. The vulnerable pot topped €1 million nine times, almost all of them between 2014 and early 2016, and peaked near €25 million in December 2014 [2]. Later spikes of €11.2 million in October 2022 and €1.2 million in February 2023 came mostly from Ethereum [2]. Some 67.25% of the vulnerable addresses held less than €100 at their peak [2].

The authors also note that most of the funds that remain vulnerable sit at a single bitcoin address in an uncommon format. They do not publish the address or a current balance [2].

What happened when they planted bait

On April 28, 2025 they deployed honeypots — addresses designed to be vulnerable — and funded them with $3 to $5 of cryptocurrency [2]. On bitcoin, transactions trying to empty them appeared in the mempool in the same second the addresses became exposed [2]. Jacquot described it to CoinDesk as "a second or a few seconds after the bait was set up" [1]. Only the addresses that reused a nonce with the same key were attacked at once, not the ones that required a more involved recovery [2].

On May 11, 2025 another wave swept the bitcoin that had been left untouched. In at least one trap, two distinct attempts spent the same funds, a sign that more than one actor was competing. They saw no inflated fees [2]. In CoinDesk's reconstruction, one bait address was left with 5,200 satoshis; they left for a stranger, signed with the bait's own private key, and landed in a block nine minutes later [1].

On Ethereum the drain was more methodical and took about three hours [2]. On Dogecoin, Dash and Litecoin the authors see no sign that anyone is watching closely [2]. A cross-chain bait between bitcoin and Ethereum was still untouched as of December 2025 [2].

What we saw on our node

The Liège dataset ends on June 30, 2024, around block 850,160 (23:59 UTC that day). We have not repeated their full historical scan. We ran two bounded, read-only passes on our full node on November 20, 2026, with the chain tip at block 976,751 (3:50 p.m. UTC). The live height is on Network and mining. We looked for repeated R values in ECDSA signatures. We did not try to recover private keys.

The first window covers 1,008 blocks from late 2014, heights 335,000 through 336,007, from 8:25 p.m. on December 19 to 3:57 p.m. on December 26, Madrid time. In that interval, close to the exposure peak described in the paper, we saw 1,655,475 ECDSA signatures. No public key signed two different inputs with the same R. A single R value appeared in two signatures from different keys, which by itself is not enough to recover a key.

The second window covers the last 10,080 blocks, heights 966,672 through 976,751, from 11:36 a.m. UTC on September 12, 2026 to 3:50 p.m. UTC on November 20. In that range we saw 70,233,613 ECDSA signatures. 55 R values appeared in more than one signature, and 3 public keys signed at least two different inputs with the same R. Those three cases are in blocks dated November 10, 13 and 17, 2026. A scan of the unspent-output set at 17:03 UTC on November 20 found no balance in the usual forms of those three keys (P2PKH, P2WPKH and P2SH-P2WPKH): 0 bitcoin. We are not publishing the addresses.

That result does not contradict the paper: an address can become vulnerable in one block and reach its peak balance later, and 1,008 blocks is a narrow slice. What it does show is that, in the ranges we covered, same-key nonce reuse is not a frequent event. We have not scanned the gap between July 2024 and September 12, 2026.

What you can do

Modern wallets do not pick a nonce at random. RFC 6979 specifies a deterministic procedure from the key and the message, which should stop the same nonce from ever being reused [2][3]. The bug shows up in old signing software, in poorly seeded random generators — the 2013 Android flaw is the usual example — or, rarely, in nonces typed by hand [2].

If you sign bitcoin, use a current wallet and stay away from custom or abandoned signing tools. If one of your addresses has signed twice with the same nonce, that key is no longer secret. Create a new address with a modern wallet and move the funds there. Sending another transaction from the same key is not enough.

If what worries you is a public key that is already visible, rather than this nonce mistake, the counts and the migration proposals are in Bunker mode: exposed keys.

Sources

Source type in parentheses. Accessed November 20, 2026. Times in UTC unless noted.

  1. Shaurya Malwa, "Bitcoin bots pounced within seconds when researchers exposed their wallet keys", CoinDesk, November 20, 2026, 8:30 a.m. EST. (Secondary, news outlet)
  2. Vincent Jacquot and Benoît Donnet, "Short Paper: Oops… I Did It Again. I Reused my Nonce", University of Liège / Financial Cryptography and Data Security; PDF. Data through June 30, 2024; prices as of November 25, 2024; honeypots on April 28, 2025. (Primary, academic paper)
  3. T. Pornin, RFC 6979, Deterministic Usage of DSA and ECDSA, IETF, August 2013. (Primary, standard)
  4. NodeWitness, scan of ECDSA signature R values on our full node, November 20, 2026, ranges 335,000–336,007 and 966,672–976,751 (tip 976,751, 15:50 UTC). (Primary, own data)

Bitcoin's signature scheme still holds. What does not hold is a wallet that reuses a number meant to be used once. On an open network, that mistake does not wait: in the 2025 honeypots, the next step arrived in a second.

More in Network and mining

See all 21 articles in Network and mining →

Last updated: November 20, 2026