Anthropic recently released a report analyzing the performance of its advanced models Cloud Oops 4.5, Cloud Sont 4.5, and GPT-5. The evaluation was based on the SCONE-bench benchmark, which includes 405 real-world smart contracts from 2020 to 2025 that had been subject to attacks. These models identified security flaws in the contracts amounting to approximately $4.6 million in losses, particularly after the latest data from March 2025. The report also included a significant test of 2,849 newly introduced contracts that showed no known vulnerabilities. During this test, Cloud Sont 4.5 and GPT-5 each discovered two new zero-day vulnerabilities, potentially causing losses totaling $3,694. Notably, the API usage cost for GPT-5 was $3,476. Anthropic, a leading artificial intelligence company, focuses on identifying and addressing various digital and smart contract security risks through advanced models. Smart contracts, which are automated agreements based on blockchain technology, facilitate cryptocurrency and digital asset transactions but can have vulnerabilities leading to financial losses. The report emphasizes the necessity of using advanced AI tools to enhance smart contract security and reduce potential attacks and financial damage. Further research and development will be essential to increase trust and protection in the blockchain and cryptocurrency sectors.
Source: binance
