SlowMist Annual Security Report: Security incidents throughout the year resulted in approximately $2.935 billion in losses, with Generative AI frequently used as an attack tool

BlockBeats News, December 30, according to SlowMist Blockchain's incomplete statistics, there were a total of 200 security incidents throughout the year, resulting in a loss of approximately $2.935 billion. Compared to 2024 (410 incidents, loss of approximately $2.013 billion), although the number of incidents has significantly decreased, the loss amount has increased by about 46% year-on-year.

From the ecological distribution perspective, Ethereum remains the most frequently attacked and severely affected ecosystem, with an annual loss of approximately $254 million, significantly ahead in proportion; BSC follows closely behind, with related losses of about $21.93 million; Solana ranks third, with an annual loss of approximately $17.45 million.

Divided by project track, DeFi projects are the most commonly attacked area: in 2025, there were a total of 126 security incidents, accounting for approximately 63% of the total for the year, resulting in a loss of approximately $649 million, a decrease of about 37% compared to 2024 (339 incidents, loss of $1.029 billion). There were only 12 exchange platform incidents, but they resulted in a significant loss of $1.809 billion, with Bybit alone losing approximately $1.46 billion in a single incident, making it the most serious incident of the year.

Looking at the reasons for the attacks, contract vulnerabilities are the main cause, with a total of 61 cases; X account breaches follow closely behind, with a total of 48 cases.

The report points out that with the accelerated popularization of Generative AI in the past two years, attackers have also begun to incorporate it into their fraud and attack chains. Compared to traditional tools, the ability of AI in text, voice synthesis, image, and video generation has significantly reduced the cost of fraud. Attacks no longer rely on rough wording or obviously abnormal behavior but instead through highly realistic content, coherent interactions, and precise object selection, make it more difficult for victims to perceive risks at a psychological level.