Loading prices…
🩸BEARISH

HAWK post-quantum signature broken: Claude AI halves key security

Current bitcoin and ether signatures are untouched, but the result lands while BIP-360 and BIP-361 argue the migration window is closing because classical cryptanalysis keeps improving by orders of…

HAWK post-quantum signature broken: Claude AI halves key security
HAWK post-quantum signature broken: Claude AI halves key security
HAWK post-quantum signature broken: Claude AI halves key security
HAWK post-quantum signature broken: Claude AI halves key security

Anthropic's Claude Mythos Preview model uncovered an attack that effectively halves the smallest key strength of HAWK, a proposed post-quantum digital-signature scheme that had survived two years and two rounds of expert review. The run took about 60 hours and roughly $100,000 in compute, and slashed the work needed to break HAWK's smallest parameter set from around 2^64 operations down to 2^38. Larger keys remain impractical to crack, but doubling them removes most of what made HAWK attractive in the first place.

Why it matters

HAWK is not deployed anywhere and is not on Bitcoin's migration list. Bitcoin and Ether today rely on elliptic-curve signatures, which the attack did not target. But BIP-360, the proposal for quantum-resistant Bitcoin addresses, leans on NIST-standardized algorithms with deliberate redundancies precisely so that if one candidate is later broken, others remain. Anthropic's result is the first public demonstration that an AI model can produce a double-digit-order cryptanalytic improvement on a leading post-quantum signature candidate in days, against a scheme that had already cleared two human review rounds.

Market impact

The near-term price read is muted. No live signature scheme used in Bitcoin or Ether was touched, and HAWK was never a candidate the network would adopt. The strategic read is sharper: BIP-361 argues the migration window is closing because classical cryptanalytic attacks are improving by up to 20-fold, and Anthropic's HAWK result aligns with that trend at a scale no human-led effort has matched in the same timeframe. The second result in the same release improved attacks on a weakened AES variant by 200x to 800x, while improvements against the Poseidon hash function that underpins many zero-knowledge rollups came in under 10x, leaving ZK-based systems relatively safer for now. Watch for NIST's next standardization round and any acceleration in BIP-360 and BIP-361 timelines.

Related tokens
$BTC $ETH

Frequently asked questions

  1. Does Anthropic's HAWK attack affect Bitcoin or Ether today?

    No. Bitcoin and Ether both use elliptic-curve signatures, which the attack did not target. HAWK is also not one of the schemes Bitcoin's BIP-360 migration plan would adopt, and the scheme is not deployed anywhere yet.

  2. What did Anthropic's Claude actually do to HAWK?

    Claude Mythos Preview found an attack that cut the work to break HAWK's smallest parameter set from roughly 2^64 operations down to about 2^38. Larger keys remain hard to crack, but doubling them removes most of what made HAWK attractive.

  3. Why is the HAWK result relevant to Bitcoin's quantum plan?

    BIP-360 and BIP-361 argue the migration window is closing because classical cryptanalytic attacks are improving by up to 20-fold. Anthropic's HAWK result fits that trend, and was produced by an AI in 60 hours rather than by a multi-year human review process.

  4. Is HAWK deployed anywhere in production?

    No. HAWK is a candidate under review to replace the digital signatures that protect web payments and online banking. It had survived two years and two rounds of expert human review before Anthropic disclosed the attack to its authors in June.

  5. What about zero-knowledge systems like Poseidon?

    Anthropic said its model produced only sub-10x improvements against Poseidon, the hash function underpinning many zero-knowledge rollups and privacy protocols. ZK-based systems look relatively safer than signature schemes from this release.

Source attribution
Aggregated from CoinDesk · Verified · Last refreshed 53m ago
Open original →