Ethereum News: In a post dated July 9, 2026, authored by Nikos Paksifanis, the Ethereum Foundation’s Protocol Security Team published a detailed account of running coordinated AI agents against Ethereum’s underlying protocol code, including systems software, cryptographic libraries, and contracts, and the main finding was the methodology, not just the vulnerability they uncovered.
Customers found a real bug: a remotely triggerable panic in the gossip layer of libp2p, the peer-to-peer substrate on which all Ethereum consensus clients rely, which has now been patched and publicly disclosed as CVE-2026-34219. But Paxivanis views this revelation as secondary to a more sustainable view of where security research time actually goes when customers enter the pipeline.
explores: The next crypto to explode in Q3
Ethereum News: The bottleneck has shifted, not gone
The central argument of this post is precise: AI agents are search tools, not oracles, and the work they create is not generation but sorting. As Paksvanis directly stated in the post, “AI has not replaced the security researcher.
He moved the work. Time that used to go into coming up with hypotheses and chasing them now goes into judging them extensively, including building the oracle, running triage, maintaining a list of known issues, and dealing with detection.
The team runs multiple agents in parallel against a single target, coordinating through shared state in version control rather than a centralized process, an approach followed in Anthropic’s published book on building a C compiler with a fleet of agents. The roles emerge from the work itself: Recon turns the attack surface into testable hypotheses; Hunting traces code paths and builds reproducers; Gap Filling tracks coverage and puts the next batch on hold; The validation process re-examines each candidate independently and makes an accept or reject call.
The protocol security team directs AI agents to the Ethereum protocol code. Our original idea wasn’t about finding errors, it was about sorting.
Below are field notes from the work.https://t.co/HVtc8XcrJK
— Ethereum Foundation (@ethereumfndn) July 9, 2026
The admission bar is strict. A filter does not become a hit until a self-contained artifact reproduces the failure against real shipping code and runs it for someone who did not write it.
The post identifies three frequent false positives that the reproducibility requirement filters out: panics that only appear in a debug build; A reproducing device that constructs an internal value that an attacker-controlled input path could never produce; And prove formal verification that is trivially satisfied no matter what the underlying code does. “What’s new is the size,” Paksifanis points out. “The agent writes the useless copy just as quickly as the real copy, and with the same confidence.”
discovers: Best coins to buy in 2026
AI agents in security: what they do well and where they mislead
The ability of the mail maps agent is frankly extraordinary. Agents effectively read the specification and code together, identify and verify true constants, and copy the draft from a one-line idea.
They mislead communication chains that appear reachable but are not, manipulate the success check to produce a statement for the wrong reason, inflate severity to match the language of dramatic writing, and, most importantly, errors that extend over a valid chain of steps where only the order is wrong.
For this last category, Paksvanis sees the agent’s role as suggesting which sequences are worth running through a stateful testing tool, not replacing them.
The post credits Stanisław Forte’s “rough bounds” framework: a model that recovers an entire exploit chain on one codebase can fail to trace the underlying data flow on another, so there’s no single good result that indicates the next result will hold.
A great blog post for security researchers.
TL:DR
– Running code-intensive AI agents can introduce several vulnerabilities.
– The product is sorted. The bottleneck now is expert human judgment. https://t.co/uKDqjz3DJ4— Cotabe.eth (@Cotabe_M) July 9, 2026
Each candidate is independently screened regardless of past performance. Parallel industry work at Anthropic’s Frontier Red Team and Cloudflare has converged on the same architecture, resurvey, parallel hunting, independent verification, and data deduplication, which the publication treats as evidence that the method is stable even as tools change rapidly.
This is not just an article about deploying AI in security workflows. It is a structural argument about where human judgment remains non-negotiable: not in generating hypotheses, but in determining what counts as evidence, what constitutes a duplicate of a known case, and what is revealed and when.
the The Ethereum Foundation’s organizational structure, as reported by CoinSpeaker,Giving this argument operational weight, the team needs a ,pipeline to scale governance, not just throughput. As Paksvanis concludes: “Ignoring that is what ends up sending the wrong message, ‘It’s okay.’”
Disclaimer: Coinspeaker is committed to providing unbiased and transparent reporting. This article aims to provide accurate and timely information but should not be considered financial or investment advice. Since market conditions can change rapidly, we encourage you to verify the information yourself and consult with a professional before making any decisions based on this content.
Neil is a professional cryptocurrency content writer with years of experience. He has written for numerous cryptocurrency websites to report breaking news, and has been hired by all kinds of cryptocurrency projects, to create content that will increase their exposure and attract more potential investors.
