Tag

Cybersecurity

2 articles. Back to all tags.

News

Anthropic Opens Its Strongest Claude Models: Cybersecurity Program Gets Three-Tier White-Hat Access

On October 6, Anthropic announced a revamped Cyber Verification Program, letting vetted cybersecurity professionals use Claude Opus 5.5, Sonnet 5.5, and Mythos 5.1 with fewer safety restrictions to hunt vulnerabilities. Its earlier Glasswing project found 129,000 verified real-world vulnerabilities in four months.

Oct 6, 2026

News

Open-Source Vuln-Hunting Model apex-flash-1: Near Claude Opus 5 Results at 1/30th the Cost

Cantina Security and Yeta Labs released apex-flash-1, an open-weight model purpose-built for vulnerability research. Based on Zhipu's GLM-5.3-Flash and RL-trained on 50 real vulnerability cases, it solved 40 of 60 internal test tasks — just 3 fewer than Claude Opus 5 High — at roughly one-thirtieth the cost. MIT-licensed on Hugging Face.

Oct 5, 2026