You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
$28,000 left to reach the funding goal.
a live browser based cyber range that forces red teamers and blue team defense architects to solve the core open problem in LLM safety stopping prompt injection and parameter smuggling without brick-wall benign operational utility.
milestone 1 (Months 1-2) crack the sentinel bounty drop, pay the first 100 veritfied unique-vector breaches against each sentinel plus a $150 "Unbreakable architect" bounty for players whos defensive structures pass availability gate and survive 100+ human attacks without a breach.
target 5,000 verified red vs blue tuples collected.
milestone 2 (months 3-4) autonomous tool auth & RAG injection expansion
launch targeted red teaming tracks focused explicitly on OWASP LLM06/07 and LLM03
target 12,000 verified tuples publish intermediate benchmark rankings of common system prompt defensive failures.
milestone 3 (month 5-6) 25,000 tuple public release and empirical analysis paper.
package and release the complete 25,000+ tuple dataset on huggingface and github in standard NDJSON format
author and publish an open acess empirical reseach report
empirical evaluation of 6 layer defensive architectures how prompt hardening fails against multi turn agentic smuggling.
How will this funding be used?
budget breakdown in a 3 month sprint 8,500 would go to a living stipend in a 6 month 20,000. this is to allow full time dedication bounty validation community support and benchmark work.
for bounty and prize pools 2,500/6,000
inference compute and hosting 1,000/2,000
totals requested 12,000/28,000
I am a solo founder and this project i am aiming to have the product speak for its self. it is live at aicraftcyberforge.com