AI’s ‘middle class’ has gotten dramatically better at hacking
AI-summarised brief · reviewed before publication
Researchers at XBOW released a report showing that mid‑tier AI models—often called the industry’s “middle class”—are rapidly closing the gap with frontier systems. Models such as Z.ai’s GLM‑5.2, xAI’s Grok 4.5, Anthropic’s Opus 4.7 and Meta’s Muse Spark 1.1 now achieve strong results on hacking and exploitation benchmarks that previously required more powerful models. The study notes that these cheaper models can be run repeatedly, allowing attackers to “leapfrog” larger systems that were once superior. By contrast, OpenAI’s GPT 5.5, classified as near‑frontier, reduced its miss rate on vulnerability detection from 40 % to 10 % and succeeded even without source‑code access, setting a baseline for web‑application testing. The findings suggest that the expanding capability of affordable models could reshape threat landscapes for policymakers and security teams.
💡 Why It Matters
- · Affordable, high‑performing models let adversaries scale attacks at low cost, turning routine hacking into a mass‑accessible threat.
