Will AI-driven smart contract exploits exceed $1 billion in total losses by June 30, 2026?

AI agents are rapidly transforming blockchain security risks. A new Anthropic report shows that advanced autonomous models — including Claude Opus 4.5, Sonnet 4.5 and GPT-5 — successfully exploited 17 of 34 newly deployed contracts in simulation, stealing $4.5M in mock funds. Across 405 historical contracts, the agents generated $550M in simulated exploit revenue. They also uncovered two entirely new zero-day vulnerabilities missed by audits and prior attackers. Anthropic notes that AI-driven exploit profitability has doubled every 1.3 months, while the cost of running such agents keeps dropping, enabling large-scale autonomous probing of on-chain code.