In a significant advancement for Web3 security, Savant Chat, a leading AI platform specializing in smart contracts audits, has achieved an impressive milestone by securing 6th place in a highly competitive public contest known as Sherlock. This event included numerous expert human auditors, marking a historic moment as it is the first time an AI has demonstrated performance comparable to experienced human auditors within such a demanding environment. This achievement is poised to transform the landscape of decentralized finance security.
The competitive nature of audit contests traditionally favors human participants, given their designed challenges. Professional auditors often highlight that the most difficult aspect is not merely identifying bugs but also substantiating them as true positives during the judging process. Success relies heavily on the auditors’ ability to effectively communicate their findings. Savant Chat, utilizing its Pro plan and incorporating cutting-edge technology from June 2025, overcame these obstacles with remarkable proficiency.
Compounding the significance of this accomplishment, the Symbiotic contract contest that tested Savant Chat did not include any critical or high-severity vulnerabilities, which is the domain where Savant’s AI is typically optimized for peak performance. AI tools often identify medium-severity vulnerabilities but do so with varying degrees of success. The fact that Savant Chat excelled in a contest focusing on less severe issues underscores its advanced analytical abilities and adaptability.
To achieve this, Savant Chat deployed a meticulous operational methodology enriched with deep analysis, threat vector generation, and the formulation of complex vulnerability hypotheses. For the crucial task of writing Proof-of-Concept (PoC) code to validate findings, the team wisely partnered with a highly regarded open-source Software Engineering (SWE) agent. This strategy maximized the efficacy of Savant’s core insights while ensuring validation of its findings.
The economic ramifications of this achievement are profound. Traditionally, resource expenditure during audit contests has been skewed heavily towards hypothesis generation, consuming about 85% of resources, while verification takes up only 5%. Although AI technologies generate slightly more noise compared to average human auditors, the costs associated with AI-driven hypothesis generation are substantially lower. This demonstrates that a model combining AI-driven hypothesis generation with human verification can be both effective and economically feasible, promoting broader security coverage across the entire Web3 ecosystem.
While AI presents considerable advantages, it is crucial to acknowledge its limitations. In specific instances, human auditors may outperform AI in identifying particular types of errors. Nevertheless, Savant Chat brings a unique perspective on identifying bugs that allows it to detect vulnerabilities that human experts might miss. Notably, it has been reported that 80% of zero-day exploits submitted to the Savant team were successfully identified and mitigated by Savant Chat. This underlines its complementary capabilities, highlighting that the integration of AI could have prevented a significant number of associated hacks, establishing AI as an essential enhancement rather than a replacement for existing security measures.
As we look ahead, the narrative of Web3 security is not just one of arrival; it’s a narrative of rapid acceleration. Savant Chat continues to reinforce its position as an integral part of the future, ensuring the security of blockchain projects through its AI capabilities that operate continuously to detect advanced threats and deliver efficient auditing solutions.
In its mission, Savant Chat combines cutting-edge technology with robust security measures to protect the evolving landscape of decentralized finance.
