The Ethereum Foundation has disclosed that a coordinated team of AI agents successfully discovered a remotely triggerable vulnerability in the libp2p's Gossipsub networking layer – a core component used by the blockchain’s consensus clients. The bug, which could have become a serious threat, was patched before any damage occurred. The Protocol Security team described the process as a major proof-of-concept for AI-assisted auditing, though they cautioned that filtering genuine issues from an overwhelming flood of false positives remains a key challenge.
Just as the foundation published its findings, another powerful signal emerged from the venture capital world. Paradigm, one of crypto’s largest VC firms, announced a new $1.2 billion fund dedicated to “frontier technology startups,” explicitly including artificial intelligence, robotics, and crypto. This marks a strategic expansion beyond digital assets alone, with the firm already backing non-crypto companies like Zipline and True Anomaly. Paradigm co-founder Matt Huang has long argued that AI and crypto will have “plenty of overlap,” and the new fund – alongside the firm’s open-source smart contract security tool EVMbench, developed with OpenAI – shows that overlap becoming concrete.
However, the same AI capabilities that are being harnessed for defense are also empowering attackers. Security researchers from CertiK and SlowMist have warned that automated AI tools are making it cheaper and faster to scan thousands of contracts, identify dormant loopholes, and launch attacks. In the first half of 2026 alone, hackers stole $1.32 billion from crypto protocols, and while the total dollar losses may have dipped in some reports, the number of attacks increased by 50%. A December 2025 study by Anthropic showed that AI agents could find exploitable flaws in newly launched smart contracts and generate attack code for as little as $3,476 in computing costs, targeting contracts worth roughly the same amount – proving that profitable AI-powered attacks are already feasible.
The dual narrative is clear: AI is simultaneously becoming crypto’s greatest security asset and its most urgent threat. The Ethereum Foundation’s bug discovery, Paradigm’s $1.2 billion commitment, and the rise of AI-native auditing tools like EVMbench all point toward a future where continuous AI probing becomes standard practice. Yet, as the foundation itself stressed, every serious finding still requires careful human review. One-time audits may no longer suffice for legacy smart contracts, and the industry is racing to build defenses at the same pace as the offense evolves.
