Over the previous two years, the decentralized finance ecosystem has proven quantifiable enchancment in its resilience towards direct sensible contract assaults. Knowledge from corporations equivalent to Immunefi and SlowMist signifies that losses from DeFi‑particular exploits fell from a peak of $2.62 billion in 2022 to roughly $680 million in 2025, a 74% discount.
The median loss per incident dropped from $6 million to $1.5 million over the identical interval. Danger classes that dominated the panorama in earlier years — equivalent to cross‑chain bridge assaults and flash mortgage manipulations — have gone from accounting for 73% and 54% of losses, respectively, to residual values of three% and fewer than 1%.
This enchancment is partly attributable to the rising adoption of AI‑primarily based instruments for code auditing and vulnerability detection. Frameworks equivalent to SimSecLLM, LLM‑SmartAudit, and SmartProof have demonstrated the flexibility to scale static and dynamic evaluation of sensible contracts, decreasing publicity home windows. Nevertheless, claiming that AI is driving a “new, extra superior, and safer period” for crypto could be, at greatest, an incomplete — and probably harmful — studying of the present second.
The opinion I maintain, primarily based on aggregated proof from the final yr and a half, is as follows: the sector is successful the technical battle over code, however shedding floor on the operational and human entrance. And AI, removed from being a one‑sided resolution, acts as a symmetric accelerator of capabilities for each defenders and attackers.
Complete crypto hacking losses contradict the narrative of a broad enchancment. In accordance with SlowMist’s 2025 annual report, the complete worth stolen from crypto hacks elevated by 46% in comparison with 2024, reaching $2.935 billion. This improve occurred in parallel with a 37% yr‑over‑yr decline in losses particularly attributable to DeFi.
The rationalization for this obvious paradox lies within the origin of the incidents: the majority of 2025 losses did not stem from sensible contract flaws, however from Web2‑fashion operational errors — password leaks, social engineering assaults, personal key compromises in poorly managed environments, and vulnerabilities in centralized trade infrastructures.
The most consultant case of this development is the 2025 Bybit assault, with a lack of $1.5 billion — an incident that, on account of its magnitude, can’t be ignored, however as a result of it doesn’t correspond to a DeFi protocol, it falls outdoors probably the most regularly cited sectoral statistics. This measurement bias is problematic. When the sector celebrates a discount in DeFi losses, it omits the truth that the assault vector has shifted towards extra fragile parts of the custody chain: human operators, entrance‑ends, and misconfigured wallets.
Synthetic intelligence is exacerbating this asymmetry
An evaluation of coverage violations involving generative fashions discovered that 67% of instances concerned malware preparation or vulnerability reconnaissance on DeFi protocols. Researchers at Binance Analysis have documented that, in managed environments, AI is twice as efficient at exploitation as it’s at detection. Moreover, roughly 60% of all inflows to scammer wallets in 2025 originated from AI‑powered schemes, together with deepfakes utilized in social engineering and the automated technology of malicious contracts that seem professional.
Dealing with this situation, a purely technical posture is inadequate. DeFi protocol safety groups have internalized the teachings of 2022: formal audits, bug bounty applications, and AI‑pushed assault simulations are actually commonplace. However the subsequent degree of threat is not in a contract’s bytecode — it’s within the interface the person indicators, the server internet hosting the entrance‑finish, or the personal key saved in a textual content file on a company desktop.
The problem for the sector in 2026 and past is not solely technical however structural. Losses within the first half of 2026 already exceed $800 million in DeFi, suggesting that the downward development shouldn’t be monolithic and that attackers are adapting sooner than many groups replace their defenses. AI, on this context, have to be understood as a threat administration device, not in its place for operational self-discipline.
Protocols that prioritize auditing automation with out reinforcing the safety of their off‑chain environments will stay uncovered to assault vectors that no LLM can mitigate by itself.
The assertion that DeFi hack losses have fallen is factually right. The assertion that AI is resulting in a brand new period of superior safety is conceptually true provided that three nuances are added: first, that the enchancment is concentrated in a single section of the ecosystem (code, not operations); second, that AI empowers each attacker and defender, with the present benefit leaning towards the attacker when it comes to exploitation effectiveness; and third, that the related metric for the tip person is not mixture DeFi loss, however complete crypto hacking losses — which proceed to rise.
The sector would do nicely to have a good time with warning and act with urgency on what is definitely failing: the human layer.
