You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
I'd like to inform you about aicraftcyberforge.com. It is a live game used as an educational resource for red-teaming and AI safety. The platform's main benefit is a five-act, 35-mission training program that gives developers, students, and the general public resources and constraints for addressing autonomous multi-agent swarms. This platform is based on a "probe, break, defend, verify" process to resolve adversarial agent configurations. Instead of using multiple-choice questions, players get hands-on experience working through each step of the process. What are the objectives of this project? How will you achieve them?
Goals
There are two goals. First, raise the bar for AI safety standards through red-team gameplay. Second, create a training platform for students and developers who want red-team practice. The unique value proposition is an active “probe, break, defend, verify” cycle of adversarial agent pairings, rather than a static code snippet database. Players can learn how to use semantic firewalls, softmax clamps, air-gapped context windows, execution circuit breakers, and other "real-world" defenses to secure systems against jailbreaks, indirect information injection, rogue agent loops, and other harmful behaviors. We can use numerous concrete KPIs to measure success. We plan to onboard at least 10,000 unique users onto the platform in year 1. Assist with >500 completed scenarios per month. Have users test 75+ different defensive tactics in the wild LLM cage match. We can also track other metrics, like user mission ratings, time to complete scenarios, and the number of teaming strategies the community contributes. What will this funding be used for?
Tier 1: $2,500 – $4,500
The funds will let learners access the resource free of charge by supporting the computing and bandwidth needed for more than 10,000 guest sessions, increasing both the project's reach and its training impact.
It would also enable the next stage of the curriculum through 10 new scenario modules that address emerging vulnerabilities, extending training coverage to include misuse of MCP tools, multi-turn system prompt extraction, and multi-agent consensus poisoning.
This would also provide ready-made workshop materials for university AI safety student groups, enabling them to organize practical training sessions and thus expand the project's outreach.
The funds would establish an Open Community Adversarial Arena. This public sandbox lets researchers and developers submit defensive prompts for live automated stress testing, permitting faster validation of defenses and broader use of the results.
It would also include an Open-Source Evaluation Harness: refine the automated evaluation runner, document it, and make it available as open source, thereby allowing other research laboratories to benchmark defensive prompts on their own local systems, reproduce the results, and apply them more consistently.
It would also include a Collegiate Red-Teaming Tournament: an online red-teaming hackathon offering small cash bounties for new jailbreak discoveries and defensive designs, producing new testing ideas and practical defensive measures.
Adding another programmer will reduce dependence on a single person, help meet the project's milestone deadlines, and ensure development continues for clients.
Because I work in web architecture and applied AI development, I focus on delivering real-world results by leveraging CyberForge's main value proposition: a fully functional 3D engine, custom audio, and a playable first act available at aicraftcyberforge.com. In past projects, I have led the development of interactive cybersecurity training platforms for university courses and advised research teams on secure multi-agent simulations. I have also worked with AI safety organizations to create proof-of-concept defenses for antagonistic scenarios. All the code, scenario logic, and security constraints are designed to be used right away and to ensure a seamless onboarding process, including local storage so offline play is possible. I intend to grow the team in the future.
Since this project is one I am genuinely passionate about, the greatest risk is being unable to cover server costs, which could force the platform to shut down, restrict access, reduce the impact of the training, and hinder continued use. To ensure the project's long-term sustainability, I plan to partner with academic institutions and industry organizations willing to sponsor workshops or provide institutional support in return for early access to modules or analytics. At the same time, I will look into offering premium features such as advanced scenario packs or professional certification tracks, while keeping the basic training free for everyone. These measures aim to secure ongoing funding and give stakeholders confidence in the project's viability after the initial seed funding runs out.
If our outreach doesn't succeed, progress in both data collection and AI safety improvements will slow, reducing the project's speed and impact; consistent outreach is essential to maintain momentum and results.
Another possible risk is that I become ill or that something unexpected happens that stops me from working on the project. In that situation, the project might lose continuity, and a portion of the funding would have to be used to hire a similarly dedicated programmer to keep delivery going and ensure the work continues.
At the moment, I have not obtained any financing since the project is still in its early phase. In the future, however, I intend to seek further funding and establish collaborations with universities, AI safety organizations, and industrial sponsors to support continued development, expand the platform's capabilities, and maintain long-term sustainability.
There are no bids on this project.