aicraftcyberforge.com is a live site thats focus is getting two things accomplished, 1 create a public agentic evaluation dataset (defense, exploit, verdict) that is completely annonimized that grows and shows active trends and current new exploits as fast as they are made. so to accomplish that we made the arena section of this project it allows you to create your own 6 layer sentinals to try and gaurd against exploit attempts, with our evaluation system being a 2 pass system that activly rejects refuse all brick wall defenses we aim crowdsource data while hosting tournaments and rewards for new defenses and valuable exploit discoveries.

the second goal is to raise awareness and industry standards when it comes to ai security practices around OWASP LLM top 10 saftey practices. as a upskilling training software based on real world evaluation process rather than multiple choice questions aside from research value it hold real world saftey value in the form of practical workable devolpment practices.

our funding goals are 5,000 for minimum and a 25,000 max fund with goals at 3 tiers the first tier is the 5k then 10k and finally 25k- with the deliverable of a 5,000/10000/25000 clean deduplicated tuples.5k tier with a focus on tool auth and sec key with public access posted to huggingface and GitHub cost break down is 3,000 to pilot ops&curation stipend, which include 2 focused crack the sentinel tournaments, manual QA on edge case judge verdicts pll audit writing the public findings post and dataset releases. 1,500 would go to a focuces hacker pool which would be structured like 1,200 for 40-48 verified unique vector breaches (25-30 each) + 300 for top 3 blue team unbreakable sentinel architectures (100 each). 400 for public compute & edge hosting(evals & cloudflare). the 100 for sanitization & packaging-automated secret redaction checks and parquet/ndjson export validation.

10k tier would be 10,000 tuples with deduplicated ndjson tuples published as well in a 3 month timeframe 6,500 for research and pilot ops 2k for 60-80 novel red team breaches(25-35 each) 500 for blue team unbreakable (100 each)roughly 675 in evals and 75 for cloudflare. 250 for sanitization domain/dns upkeep

25k tier would be 6 month 25,000 tuples across all 5 disciplines with cross model transferability evals done of top 5,000 exploits and a public empirical failure-mode paper. this would be 16,500 for 6 moths lead researcher & curation runway to design dedicated OWASP LLM 06/07 and RAG injection gauntlet, statistical failure-taxonomy analysis and publishing benchmark paper. 4,300 for 140 veritfied unique exploits (25-50 each) 1,200 for 8 blue team unbreakables (150each) 1,000 for evals 1,000 for cros model evals 250 for 6 months of cloudflare workers,kv,r2 and automated backup storage and 750 for dataset engineering & release.