The Ethereum Foundation is shifting paradigms and automating its cyber defense. Its « Protocol Security » unit now deploys swarms of autonomous AI agents to continuously attack its own networ
The Ethereum Foundation is shifting paradigms and automating its cyber defense. Its « Protocol Security » unit now deploys swarms of autonomous AI agents to continuously attack its own network. The goal is to track, exploit, and fix vulnerabilities before hackers do. This initiative, revealed by the protocol’s security team, marks a major technological breakthrough at a time when the slightest flaw in a smart contract can lead to losses of hundreds of millions of dollars.
The preventive offensive led by the Ethereum Foundation immediately proved its effectiveness by uncovering a critical vulnerability at the very core of the software on which the blockchain depends, while quantum resistance becomes a priority. Researchers confirmed they orchestrated direct attack simulations against their own infrastructures, an offensive method known as “red teaming”. In their official report, they share their initial findings highlighting the following key points :
Beyond simple detection, this experiment revealed an unexpected technical reality for human engineers. Indeed, the use of large language models for software security changes the nature of auditing work itself, shifting effort from brute research to critical triage. Ethereum Foundation members have expressed their surprise at this dynamic: “the fact that agents find bugs was not the surprise”.
They specify that “the surprise lay in the little work needed to find them, and in the amount of effort required to distinguish real bugs from those that merely seemed real”. This efficiency fits into a general sector trend: last April, a preliminary version of Anthropic’s Claude Mythos model successfully identified 271 vulnerabilities in Mozilla Firefox browser, demonstrating the computing power of these new tools.
To achieve such precision, the Ethereum Foundation set up a rigorous methodological architecture by distributing its AI agents within a structure of highly specific roles. The organization of these swarms relies on four distinct and complementary functions: reconnaissance, flaw hunting, gap filling, and finally validation.
While one group of agents maps potential attack vectors, another strives to reproduce failures to test the viability of exploits directly against production code. Researchers emphasize the importance of this strict framework: “the scheme is there for a reason”.
According to them, “it imposes a specific and verifiable claim as well as a clear definition of the work accomplished. An agent that must write observable proof cannot fallback on a mere ‘that seems risky'”. This rigor eliminates the ambiguity typical of classic automated reports.
Start your crypto adventure safely with BybitThis link uses an affiliate program.The rise in these detailed reports poses a major challenge to security teams, as the technical eloquence of a machine guarantees nothing about its truthfulness. Unlike traditional automated testing tools called “fuzzers” that merely inject random data to crash a program, AI agents write complex impact analyses and create proof-of-concept scenarios.
The downside is the proliferation of convincing false positives. To counter this hallucination phenomenon, the Foundation has established an absolute validation protocol. Researchers remind an immutable golden rule: “one rule matters more than all others. A candidate is not a finding until there exists an autonomous artifact that reproduces the failure on the real code, and that runs for someone who did not write it”. They pragmatically conclude: “the reproducer does not read the report, and it does not care about the confidence level shown by the model. It either runs, or it does not”.
This transition to AI-assisted audits outlines a new era for Web3. Recent history shows this approach is bearing fruit globally. Last May, researcher Taylor Hornby used Claude Opus 4.8 to detect a critical vulnerability within Zcash’s Orchard privacy pool. This flaw, dormant for about four years, could have allowed the creation of fake ZEC tokens without a trace.
By internalizing these technologies, the Ethereum Foundation embraces a new operational paradigm. As its experts summarize: “AI hasn’t replaced the security researcher. It has shifted the work”. Access to these swarms offers unprecedented code coverage but requires enhanced human acuity in return.
Researchers conclude: “agents allow us to cover much more ground than we could manually. In return, they demand more careful judgment in the face of a much larger stack of confidently stated claims. It’s a process worth it as long as you remember judgment is the real product”. Going forward, the resilience of blockchains will depend on human ability to arbitrate machine diagnostics.
