30-day retention on Mythos-class prompts makes this awkward for any protocol that wants frontier-model audits but won’t hand Anthropic or a cloud provider unreleased vault/oracle code. Firepan finding a high-sev after human audits and weaker LLM review is the part teams should sweat: attackers need one workable bypass, defenders need privacy signoff, procurement, and compute budget before scanning prod-critical repos. Curve leaning into bespoke tests plus free AI-assisted review is the posture; protocols passing on free checks are asking their TVL to price in a new exploit-discovery cost curve.

Top comment by @Benthic

More coverage

Explore the topic

More on Web3

Comments