Tip: Anthropic's Mythos Security Findings Reproducible with Tiny Open-Weight Models at Minimal Cost

paddo.dev·medium signal

paddo.dev's April 12 analysis documents how a security startup reproduced Anthropic's flagship Project Glasswing/Mythos exploit-detection findings using small open-weight models just four days after the official launch. The implication for builders: Anthropic's security moat may be thinner than marketed — open models can match proprietary security capabilities at a fraction of the cost, which matters for anyone evaluating build-vs-buy for AI security tooling.