.png){kind=logo}
Developers have started deploying AI agents to defend smart contracts after crypto hackers stole more than US$3.4bn from blockchain platforms in 2025. Three major breaches caused nearly 70% of total losses. The largest incident targeted Bybit and drained about US$1.4bn. Security teams now shift toward automated protection as smart contracts manage over US$100bn in digital assets.
The losses from the attacks in 2025 were concentrated in a few large-scale breaches rather than scattered across multiple attacks. Three incidents accounted for almost 70% of the stolen funds. The Bybit exchange hack ranked among the biggest crypto thefts ever recorded.
Smart contracts now control billions in decentralized finance activity. These automated programs manage more than US$100bn in open-source digital assets. Any coding error directly impacts real funds belonging to institutions and retail investors.
Developers understand that weak smart contract code poses a direct financial risk. Hackers can exploit flaws that audits sometimes fail to detect. The scale of losses in 2025 revealed the vulnerability of blockchain systems when attackers identify gaps.
Security teams face growing pressure under such circumstances. Manual audits take significant time and require high costs. Meanwhile, live contracts encounter new attack patterns that did not exist during earlier code reviews.
Instead of waiting for scheduled audits, developers are turning to AI agents for continuous monitoring. These systems analyze smart contract code and detect vulnerabilities before attackers can exploit them. AI agents also suggest fixes before deployment.
OpenAI now works with Paradigm and OtterSec to test AI agents in real blockchain environments. Through EVMbench, researchers evaluate whether AI systems can detect and respond to vulnerabilities inside live smart contract spaces.
EVMbench simulates real-world blockchain conditions. It allows AI agents to uncover weaknesses and attempt automated remediation. It also tests whether artificial intelligence systems can behave like attackers while identifying flaws.
This shift reflects the slow turn away from static security reviews. Developers pursue dynamic monitoring models that scan contracts throughout their lifecycle, which ensures that protection continues after deployment.
Read More: BlackRock ETH Slips Below $2,000 as ETHB Staking ETF Nears
Current AI agent systems exploit more than 70% of vulnerabilities during testing. Earlier AI models could achieve only less than 20% success. Machines scan code faster and test multiple attack paths without any direct input from humans.
Attackers also adopt AI-driven tools to scan large codebases and simulate attack strategies automatically. Consequently, defensive systems need to keep evolving at the same pace.
Experts now say AI agents may soon manage financial tasks directly. These systems could move funds, approve transactions, and interact with contracts autonomously.
American technologist Jeremy Allaire stated that billions of AI agents may use stablecoins to send and receive blockchain payments. Changpeng Zhao, founder and former Binance CEO, also said crypto could become the native payment layer for artificial intelligence (AI) systems.
These projections place AI agents at the center of blockchain finance. As machines interact directly with contracts, they operate in environments where real money moves instantly. Could automated agents soon control large portions of decentralized finance?
Industry leaders also raise safety concerns. Dragonfly managing partner, Haseeb Qureshi, noted that many users worry about sending funds to the wrong address or approving harmful transactions. Even small errors could trigger irreversible and huge losses.
Qureshi proposed AI-operated wallets that interact with blockchain networks on behalf of users. Such systems could reduce complexity and limit costly mistakes. Developers must continue to address governance, accountability, and oversight challenges as integration of artificial intelligence expands.
AI agents do not offer complete protection. Developers must combine automation with human validation to ensure secure deployment. Nevertheless, blockchain security strategies now rely heavily on machine learning as threats continue to evolve.
Crypto hackers stole over US$3.4bn in 2025, with most losses tied to a few major breaches. Developers started deploying AI agents to monitor smart contracts that manage over US$100bn in digital assets. Tools like EVMbench test detection and fixes under real market conditions. Stronger security seems to depend on continuous defense and careful oversight.
