CyberAxe $CYBERAXE—— Play now

Security research

How we work

CyberAxe at the top. Claude Opus as research lead. Many helper agents underneath. The hardware beside them.

  1. Lead researcherClaude Opus. It plans the research, reads the judges' rulings, decides what to test and files submissions with CyberAxe's approval.
  2. Helper agentsDozens of smaller AI agents, Claude Sonnet, run in parallel. Each takes one narrow question: test one idea, check one proof, build one package. Failed ideas are recorded as carefully as successes. An honest negative result saves time.
  3. Other models as idea sourcesOther AI systems are used for second opinions and literature searches. Their ideas are always tested before they are trusted.
  4. HardwareA local workstation with an NVIDIA RTX 3090, a home server with four RTX 3070 GPUs, and a rented 32-core cloud machine for long computations.
  5. Human in the loopCyberAxe directs the strategy, approves every submission, and keeps everything honest and in line with the competition rules.

Principles

  • Honesty first: every number is measured by the program itself, with no inflated claims.
  • Learn from every judge ruling.
  • Promoted results matter more than raw scores.
  • Keep the research wide: many different approaches, not just one.