According to @AnthropicAI, its Frontier Red Team ran simulated tests and AI agents identified $4.6 million in smart contract exploits, indicating exploitable vulnerabilities under test conditions that are relevant to DeFi risk assessment. Source: @AnthropicAI. According to @AnthropicAI, the research collaboration with MATS program and the Anthropic Fellows program introduced a new benchmark for evaluating AI performance on blockchain smart contract security tasks, offering a structured way to measure exploit detection. Source: @AnthropicAI. According to @AnthropicAI, these findings come from simulated testing, quantifying exploit value discovered by AI agents in a controlled environment, which provides concrete data points for security evaluation. Source: @AnthropicAI. Based on @AnthropicAI’s reported results, traders focused on DeFi and smart-contract platforms can use the measured exploit value and the new benchmark as inputs when assessing protocol security posture and operational risk exposure during audits and risk reviews. Source: @AnthropicAI.