The crypto industry is entering a new era of cybersecurity, where traditional one-time smart contract audits are rapidly losing relevance. My analysis of data from the first half of the year shows that AI tools are not just accelerating vulnerability discovery — they are fundamentally changing the rules of the game, rendering old code checks nearly useless.
According to my calculations based on data from leading analytical platforms, Web3 sector losses for the first six months of this year amounted to approximately $1.32 billion across 344 recorded incidents. Net damage, accounting for recovered funds, is estimated at $1.2 billion. Critically, the main attack vector is code vulnerabilities: 204 cases totaling $151.6 million. At the same time, there is a worrying trend: attackers are systematically returning to contracts older than one year, indicating the use of automated AI scanners to find defects in old code bases.
Why Old Audits Are an Illusion of Security
A smart contract can remain active for years, but during this time, bug-finding tools change, new exploitation techniques emerge, and support teams often lose interest in old protocols. The "maximum vulnerability window" does not close after launch — projects with legacy infrastructure should consider re-auditing as a regular operational requirement. The case of Zcash is particularly telling: a critical vulnerability in the Orchard pool existed from May 2022 until an emergency fix in June 2026. This bug allowed for the undetected creation of unlimited counterfeit ZEC within the private pool, and the team could not cryptographically prove whether it was exploited before the fix.
TRM Labs analysis recorded 207 hacker attacks over the half-year — a record number. However, the main damage (about 76% of stolen assets) came not from smart contract exploits, but from major infrastructure compromises: attacks on keys, signatures, and fund management systems. A separate factor is the activity of DPRK-linked groups, which stole approximately $643 million (roughly two-thirds of all stolen funds), with nearly all of this damage resulting from two April attacks against Drift Protocol and KelpDAO.
AI Agents: From 2% to 55.88% Success in One Year
Anthropic's research demonstrates an explosive growth in AI capabilities in this area. On the SCONE-bench benchmark, agents tested 405 real-world exploited contracts from 2020–2025. The result is impressive: on a subset with vulnerabilities after the model's knowledge cutoff date, success rates rose from 2% to 55.88% in one year. The total value of successfully simulated exploits increased from $5,000 to $4.6 million. The average cost of a full scan of a single contract is just $1.22 — making AI attacks accessible to any attacker.
The case of Aztec Connect is particularly telling: an attacker drained approximately $2.19 million from a private transaction solution whose support was discontinued back in 2023. This clearly demonstrates that even "dead" protocols remain at risk.
My professional opinion: The crypto industry stands on the brink of a fundamental shift in the security paradigm. One-time audits are becoming a thing of the past — they are being replaced by continuous monitoring using AI tools. Projects that fail to implement systems for constant scanning and automatic smart contract updates risk becoming easy prey for attackers armed with AI. Of particular concern is that North Korean groups have already demonstrated the ability to use these technologies for large-scale attacks at the infrastructure level.