Prophet.ai

The Problem Popular prediction markets use optimistic oracles like UMA which have several significant challenges:

Oracle Manipulation: Vulnerable to price manipulation during disputes despite "priceless contracts" Centralization: Concentrated token ownership (98% retained by team) limits decentralization Slow Dispute Resolution: Token-holder voting system creates delays for time-sensitive applications Security Economics: Relies on attack costs exceeding potential profits; vulnerable during market extremes Limited Oracle Infrastructure: Minimal architecture creates vulnerabilities when robust alternatives aren't available Case Study: Polymarket Disputed Resolution The claim "Trump declassifies Epstein list in first 100 days?" was resolved to "Yes" on Polymarket, leading to significant user disputes:

Trump declassifies Epstein list on Polymarket

Users strongly contested this resolution:

Comment 1

Comment 2

Prophet AI provides a more nuanced and appropriate response, concluding it is too early to resolve the market. Not enough substantive evidence has been released to resolve to "Yes," and the 100-day deadline has not been reached yet.

Too early claim

Case Study: Scientific Consensus Manipulation Another troubling example is the Polymarket question "Is Earth flat?" with over $76,000 in trading volume, showing a 3% chance of "YES" despite overwhelming scientific consensus.

Earth is flat market on Polymarket

Some users are clearly trolling or manipulating the market:

Comment on flat Earth market

This case demonstrates how even settled scientific questions can be distorted in prediction markets when resolvers like UMA are susceptible to manipulation. Prophet AI agents provide definitive, evidence-based resolutions that eliminate such manipulation potential:

Our Solution Prophet.ai leverages AI agents alongside blockchain infrastructure to prevent bias and ensure censorship resistance in oracle solutions for prediction markets.

Key Benefits Multi-source Verification: AI agents access Internet data through multiple sources Diverse LLM Processing: Information is processed using various LLMs in a jury-like architecture Decentralized Decision-making: Simulated agents debate for and against claims Censorship Resistance: Runs on multiple instances across TEE operators. Transparent Results: Final aggregation evaluates evidence and delivers clear resolutions The Prophet AI agent runs on multiple instances across a network of TEEs. TEEs ensure that Prophet Agents operators are running genuine Prophet instances. Users submit requests for claim verification via an on-chain smart contract, and an available operator picks up the task, executes the agent, and submits the verified result back on-chain.

Agentic Verification Mechanism The agentic mechanism used by Prophet to evaluate a claim is described below:

Prophet AI Agents Architecture

Research Phase: After receiving a claim or statement, Prophet Agents leverages two teams of simulated agents:

Blue Team: Argues in favor of the claim Red Team: Argues against the claim Each team conducts comprehensive research using multiple search providers:

Tavily Exa Serper Perplexity Twitter A team aggregator builds the strongest possible case from the evidence. Note that each aggregator need not agree with its team if the evidence is inadequate.

Evidence Evaluation: Team aggregators' statements are submitted to a final aggregator for evaluation.

Assessment Generation: The final aggregator produces a detailed report including:

Supporting evidence Confidence score (0-100%) One of the following resolutions: True: Claim is accurate based on evidence False: Claim is inaccurate based on evidence Inconclusive: Insufficient evidence to decide Depends: Claim requires additional context Too early: Evidence will become available in the future Blockchain Infrastructure Prophet utilizes an agentic workflow in a decentralized architecture that combines AI-powered research with blockchain verification.

System Participants The following parties interact in the Prophet network:

Market Creators: Submit claims to be verified along with payment Operators: Run instances of Prophet Agents in TEEs and register on the operator registry smart contract Smart Contract System: Manages verification process, operator registration, and TEE verification Verification Workflow Operator Registration: Operators run Prophet AI agent in a TEE and register on the OperatorRegistry contract Claim Submission: Market creators submit claims with payment through the smart contract Task Assignment: Tasks are assigned based on submission timestamp Task Execution: Operators run claim verification on the Prophet AI agent On-chain Verification: Assessment is submitted to the blockchain as a permanent record Payment Processing: Operators receive payment, and users receive verification results Note: Prophet only submits verification results on-chain without detailed reports due to gas considerations.