Estimating worst case frontier risks of open weight LLMs

Practical AI: Tools, Models & Frameworksfine-tuning

What changed

In this paper, we study the worst-case frontier risks of releasing gpt-oss. We introduce malicious fine-tuning (MFT), where we attempt to elicit maximum capabilities by fine-tuning gpt-oss to be as capable as possible in two domains: biology and cybersecurity.

Why it matters

A concrete addition to Practical AI: Tools, Models & Frameworks: it changes what's available to builders today rather than being general commentary.

How it compares

Related prior coverage to compare against:

  • gpt-oss-120b & gpt-oss-20b Model Card
  • gpt-oss-safeguard technical report
  • Introducing gpt-oss

Sources