ACCORDING to Dark Reading, the piece examines fears that frontier large language models such as Anthropic's Mythos and OpenAI's GPT-5.5 could drive an era of industrialised, autonomous exploitation across platforms, a notion described as a potential existential threat by some. Tara Seals notes that RunSybil CEO Ari Herbert-Voss argues defenders must adjust risk calculations to ever-accelerating AI threats, while human expertise remains crucial for validating which discovered bugs have real security impact.
The article quotes Herbert-Voss discussing how, between 2023 and 2026, the average time to move from bug discovery to exploitation dropped from five months to 10 hours, and highlights the idea that “shifting left” will become increasingly vital as organisations ship bugs at pace. It also covers his observation that LLM-based offensive improvements vary by vulnerability class, with Mythos showing massive gains on shallow bugs but only modest gains on mid-tier issues, necessitating extensive human review.
Published on 27 April 2026, the piece frames defenders as facing millions of automated signals while attackers depend on being lucky only occasionally.