The Frontier Model Forum is an industry non-profit dedicated to advancing the safe development and deployment of frontier AI
Washington D.C.
Joined October 2023
- Tweets61
- Following7
- Followers848
- Likes1
The Frontier Model Forum welcomes the emphasis on frontier responsibilities in the White House Accord announced yesterday.
We look forward to continuing to work across industry and the broader ecosystem to advance the norms, practices, and standards needed to develop frontier capabilities safely and securely.
AI agents can help defenders protect critical networks, systems, and data with greater speed and scale.
Our latest publication examines emerging defensive uses and considerations for agent deployment:
frontiermodelforum.org/issue…
The piece builds on our prior issue brief on AI for Cyber Defense from 2024:
frontiermodelforum.org/updat…
Frontier AI agents and AI-enabled biological tools hold significant promise for biological research, but their combination could also lower barriers to misuse.
Our latest issue brief examines preliminary risks and considerations for AI agents & bio tools:
frontiermodelforum.org/issue…
As frontier AI capabilities advance, it will be critical to develop evaluations for CBRN and advanced cyber risks that work across languages, regions, and cultures. Our latest issue brief highlights emerging practices for multilingual safety evaluations: frontiermodelforum.org/issue…
By combining advanced reasoning with the ability to act and call tools, frontier AI agents hold significant economic and scientific promise – yet also pose potential risks.
Our new issue brief identifies emerging practices for securing the agent stack:
frontiermodelforum.org/issue…
As frontier AI capabilities evolve, information sharing, incident reporting, and incident response will play an important role in managing frontier AI risks. Our new issue brief overviews each and highlights the FMF's ongoing information-sharing efforts:
frontiermodelforum.org/issue…
New Issue Brief: Adversarial Distillation
Although authorized distillation has legitimate use cases, adversarial distillation can enable malicious actors to extract frontier model capabilities while bypassing built-in safety measures.
Read more here: frontiermodelforum.org/issue…
As announced last March, the member firms of the FMF signed a first-of-its-kind agreement designed to facilitate voluntary information-sharing related to unique risks from frontier AI.
We are pleased to share a progress update here: frontiermodelforum.org/updat…
NEW: Technical Report on Managing Cybersecurity Risks in Frontier AI Frameworks
Building on our prior series on frontier AI frameworks, this technical report outlines emerging risk management practices in the cyber domain.
Read more: frontiermodelforum.org/techn…
🚨 NEW: Research Update on Frontier AI and Nuclear Security
Over the past year, the Frontier Model Forum partnered with nuclear security experts to conduct preliminary research.
Read our initial findings: frontiermodelforum.org/resea…
New Publication: Chain of Thought Monitorability
Our latest issue brief explores how Chain of Thought monitoring can help prevent certain types of harm, and why it shows promise as a new layer of defense for frontier AI safety and security: frontiermodelforum.org/issue…
🧵 (1 of 3) NEWS: A new cohort of 11 grantees have been awarded more than $5 million through the AI Safety Fund for projects in #Biosecurity, #Cybersecurity, & AI Agent Evaluation and Synthetic Content. frontiermodelforum.org/updat…
(2 of 3) The projects were selected from over 100 competitive proposals through a rigorous review process. Read more about them in the announcement.
🧵 NEW TECHNICAL REPORT (1 of 3)
Our latest technical report outlines practices for implementing, where appropriate, rigorous, secure, and fit-for-purpose third-party assessments.
Read more here: frontiermodelforum.org/techn…
(2 of 3) The technical report is the latest in our series on frontier AI frameworks. The first three can be found here:
Capability Assessments 👉 buff.ly/WIJj6K4
Risk Taxonomies and Thresholds 👉 buff.ly/sAnHU62
Mitigations 👉 buff.ly/FZdmOBD
🧵 NEW ISSUE BRIEF (1/5)
As frontier AI capabilities advance, it's crucial to develop robust risk management practices for AI and biology.
Our latest issue brief outlines the current landscape of AIxBIO safeguards, highlighting over 20 mitigations and related best practices.
In addition, our preliminary taxonomy of AIxBIO safety evaluations is available here:
frontiermodelforum.org/updat…