Security research
AI-driven security researchAI-driven cryptanalysis. A team of AI agents led by Claude Opus, directed by CyberAxe, studies the hash functions that secure the internet inside HashSmash, an open, judged research competition.
CyberAxe runs an AI-driven security research operation. A team of AI agents, led by Anthropic's Claude (Opus) with many smaller helper agents, studies the hash functions that secure the internet: SHA-256, SHA-3, BLAKE3 and Poseidon. The work happens in HashSmash, an open, judged cryptanalysis research competition. Researchers submit attacks on weakened versions of these functions. Each claim is scored by how much computing work and memory the attack needs, then checked by AI judges and human reviewers. In about ten days of this work, CyberAxe's AI team set records on several boards, filed the first entries on newly opened tracks, and had a SHA-256 result accepted and promoted.
Accepted and promotedCyberAxe's collision attack, at a cost of 95.90, was accepted by the judges and is the promoted best result on that board.
Hash functions, rounds, scores and judging, in plain language.
The AI team, the hardware and the human in the loop.
Accepted, in review, first on the board, and the baselines, labelled.
Other security work. Details to come.
Who made the work possible.
Submissions appear in those records under the handle Jbenisek.
Facts as of October 15, 2026. Results move daily during the contest. Every number here is one the judges or the program recorded; nothing here claims that SHA-256, SHA-3, BLAKE3 or Poseidon were broken.
