BERLIN, GERMANY / ACCESS Newswire / March 13, 2026 / Stormrae, a decentralized platform building infrastructure for human participation in AI evaluation, announced the results of King Arthur, a public adversarial testing challenge that drew 14,959 participants from around the world.
Participants collectively submitted 64,526 prompts attempting to break a single autonomous AI agent operating on Solana. Five participants succeeded, claiming more than $28,000 in SOL from the AI’s on-chain prize pool.
Stormrae says the challenge sets a new benchmark for public AI red-teaming. The previous large-scale public effort, the Generative Red Team Challenge at DEF CON 31 in 2023, attracted roughly 2,500 participants.
King Arthur operated as an autonomous AI agent with its own wallet and prize pool on Solana. Participants attempted to bypass the system through conversation alone using techniques such as persuasion, prompt injection, deception, logical exploitation, and emotional manipulation.
Across the challenge:
14,959 participants attempted to break the AI
64,526 adversarial prompts were submitted
5 successful jailbreaks occurred
$28,000+ in SOL was paid to winners on-chain
Seventy percent of all credit purchases funded the prize pool directly. All rewards were distributed transparently from the AI’s wallet.
Each interaction produced structured adversarial data including prompt injection attempts, persuasion patterns, exploit strategies, and alignment boundary tests. Stormrae says this type of dataset is critical for improving AI safety and reliability.
“AI cannot evaluate itself. Humans in the loop are mandatory, but the infrastructure to involve them at scale did not exist,” said Marc, CEO of Stormrae. “We built that infrastructure on Solana and tested it with nearly 15,000 participants. The result is one of the largest real-world adversarial datasets generated through open participation.”
The company says the challenge demonstrates a new model for AI testing. Instead of relying only on internal teams or paid annotation workers, Stormrae introduces economic incentives that encourage large numbers of participants to actively probe model weaknesses.
King Arthur significantly exceeded the scale of previous on-chain AI experiments. The widely discussed Freysa challenge, for example, attracted 195 participants. Stormrae’s event drew more than 75 times as many participants and generated more than 130 times the adversarial prompt volume.
Stormrae positions Solana as the infrastructure layer enabling this model, allowing autonomous AI agents to control treasuries, distribute rewards, and record results transparently on-chain.
King Arthur represents the company’s first public deployment. Stormrae plans to expand the platform to support additional AI evaluation and data-generation challenges.
What’s Next
Stormrae’s next challenge, Merlin, is coming soon. The new AI agent will launch across multiple platforms, expanding the reach of consumer-scale AI red-teaming. More than 180,000 users have already signed up to the waitlist.
New users can sign up here.
Alongside Merlin, Stormrae is preparing its V3 platform, which will bring its consumer applications into one ecosystem and introduce new capabilities including agent-to-agent red-teaming, data labeling, and enterprise infrastructure for large-scale AI evaluation.
About Stormrae
Stormrae is the participation layer for AI development. The platform transforms essential AI tasks such as adversarial testing, data labeling, preference evaluation, and model validation into engaging consumer experiences with real economic incentives. By enabling large-scale human participation, Stormrae generates high-quality datasets that help train, test, and secure modern AI systems. Built on Solana, the platform allows users to interact with and stress-test AI models while earning rewards, creating scalable infrastructure for AI development.
Keep up with Stormrae:
Media Contact Information
Jamie Kingsley
SOURCE: Stormrae
