Research agenda proposed: scale coding ability while scaling down cyber offense
terryyuezhuo · x · 2026-10-30
A researcher proposes two safety research directions: (1) scaling models' coding ability up while scaling their cyber-offense capability down, and (2) irreversible unlearning of offensive capabilities.
More from Safety
- GitHub Secure Open Source Fund backs 216 projects with 20B monthly downloads — 0xkarasy · 2026-10-30
- How an engineer broke the x402 agent payment protocol 20 ways, including a 10x overcharge — AI Engineer · 2026-10-30
- Leaked 'DeepSeek-cyber' tease: open DeepSeek models near frontier on cybersecurity CVE benchmarks — teortaxesTex · 2026-10-30
- Anthropic's internal model found and exploited critical bugs in Chrome, Safari, Android and more — emmanuelvivier · 2026-10-30
- AI Governance in Finance: When No One Can Explain Who Approved the Model's Decision — mikeflache · 2026-10-30
- Mistral Open-Sources 52B-Active Flagship at Half Price; AI Safety Incidents Roundup — Latent Space · 2026-10-30