Will an AI frontier lab crack a commonly used cryptography standard by end of March 2027?
Delphi’s forecast
- Yesyour pick1.8%$0.02
- No98.3%$0.98
This quote has expired. Ask again for a live price.
Forecast reasoning
Anthropic's Mythos model found attacks on HAWK (a post-quantum candidate) and reduced-round AES, but neither affects deployed systems; no frontier lab has yet cracked commonly used standards like full-strength RSA, AES, or ECC in their real-world forms despite rapid AI progress in cryptanalysis.
Settlement
Resolution criteria: - "Yes": Wins if, strictly after 2026-10-08T18:11:39.219Z and at or before 2027-03-31T23:00:00.000Z, at least one AI frontier lab publicly discloses or publishes a cryptanalytic break of a Qualifying Standard, with the result independently verifiable (e.g., via a peer-reviewed paper, a preprint, or an official lab disclosure accompanied by a working proof-of-concept or verifiable artifact). - "No": Wins if no such qualifying break is publicly disclosed and verifiable within the window. Clarifications: - "AI frontier lab" means an organisation that develops and releases state-of-the-art large-scale AI models (e.g., Anthropic, OpenAI, Google DeepMind, Meta AI, xAI, Mistral, or equivalent). - "Qualifying Standard" means a cryptographic algorithm that is (a) actively standardised and widely deployed in production systems (e.g., AES-128, AES-256, RSA-2048 or larger, ECDSA P-256 or P-384, SHA-256, SHA-3, or TLS 1.3 cipher suites), and (b) broken in its full, unmodified, production-parameter form — not a reduced-round variant, a challenge benchmark number (such as RSA Factoring Challenge entries), a simplified academic version, or a NIST candidate not yet finalised as a standard. - A "break" means a concrete attack that either recovers a secret key, forges a signature, or decrypts ciphertext without the key faster than the algorithm's claimed security level (e.g., sub-2^128 work for AES-128), demonstrated with a working implementation or a verifiable mathematical proof — not merely a theoretical speedup that remains above the security threshold.
Mar 31, 2027, 11:00 PM UTC