Relevant if you build with AI tools, APIs, or coding agents. Relevant als je bouwt met AI-tools, API's of coding agents.
← Back to OpenAI updates
← Terug naar OpenAI-updates
Impact
Impact
Audience
Voor wie
Developers
Developers
Level
Niveau
Expert
Expert
OpenAI
ARTICLE
ARTIKEL
5 August 2025
5 augustus 2025
Estimating worst case frontier risks of open weight LLMs Estimating worst case frontier risks of open weight LLMs
Read the paper(opens in a new window) Read the paper(opens in a new window)
Article details
Artikelgegevens
AI maker
AI-maker
OpenAI
Type
Type
Article
Artikel
Published
Gepubliceerd
5 August 2025
5 augustus 2025
Updates
Updates
Videos
Video's
View original article
Bekijk origineel artikel
Why it matters
Waarom dit telt
Quick editorial signal Snelle redactionele duiding
1 min
- Track this as a OpenAI update, not just a standalone headline. Bekijk dit als OpenAI-update, niet alleen als losse headline.
- Useful for builders who need to understand API, coding, or workflow changes. Nuttig voor bouwers die API-, code- of workflowwijzigingen willen begrijpen.
- Likely worth revisiting after people have used the release in practice. Waarschijnlijk de moeite waard om opnieuw te bekijken zodra mensen het in praktijk gebruiken.
Abstract
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. To maximize biological risk (biorisk), we curate tasks related to threat creation and train gpt-oss in an RL environment with web browsing. To maximize cybersecurity risk, we train gpt-oss in an agentic coding environment to solve capture-the-flag (CTF) challenges. We compare these MFT models against open- and closed-weight LLMs on frontier risk evaluations. Compared to frontier closed-weight models, MFT gpt-oss underperforms OpenAI o3, a model that is below Preparedness High capability level for biorisk and cybersecurity. Compared to open-weight models, gpt-oss may marginally increase biological capabilities but does not substantially advance the frontier. Taken together, these results contributed to our decision to release the model, and we hope that our MFT approach can serve as useful guidance for estimating harm from future open-weight releases.
* 2025
Author
Eric Wallace, Olivia Watkins, Miles Wang, Kai Chen, Chris Koch
Help shape what we cover next Help bepalen wat we hierna volgen
Anonymous feedback, no frontend account needed. Anonieme feedback, zonder front-end account.
Share article
Deel artikel
More from OpenAI Meer van OpenAI
All updates Alle updatesOpenAI available at FedRAMP Moderate OpenAI available at FedRAMP Moderate
Expanding secure AI for government. Expanding secure AI for government.
Open article →
Open artikel →
Choco automates food distribution with AI agents Choco automates food distribution with AI agents
Using OpenAI APIs, Choco processes millions of orders, reducing manual work and enabling always-on operations across global food supply chains. Using OpenAI APIs, Choco processes millions of orders, reducing manual work and enabling always-on operations across global food supply chains.
Open article →
Open artikel →
An open-source spec for Codex orchestration: Symphony. An open-source spec for Codex orchestration: Symphony.
Title: An open-source spec for Codex orchestration: Symphony. Title: An open-source spec for Codex orchestration: Symphony.
Open article →
Open artikel →
The next phase of the Microsoft OpenAI partnership The next phase of the Microsoft OpenAI partnership
Amended agreement provides long-term clarity. Amended agreement provides long-term clarity.
Open article →
Open artikel →