FMF Announces First-Of-Its-Kind Information-Sharing Agreement
The Frontier Model Forum (FMF) announced that all member firms have signed an agreement to facilitate information-sharing about threats, vulnerabilities, and capability advances unique to frontier AI.
Information-sharing has been central to the FMF's mission since its July 2023 launch, when it was tasked with establishing trusted, secure mechanisms for sharing information among companies, governments, and stakeholders regarding AI safety and risks.
Over the past year, the FMF worked with member firms' experts and legal counsel to define the scope of voluntary inter-firm information-sharing related to frontier AI safety and security, and establish necessary legal protections. This information-sharing is narrowly scoped to manage key risks to national security and public safety, including chemical, biological, radiological, and nuclear (CBRN) and advanced cyber threats.
The FMF also explored preliminary briefings of key lessons from member publications, including Anthropic's jailbreaking vulnerability disclosures and OpenAI's threat intelligence reports, where the organization facilitated responsible notification among researcher counterparts at different firms.
The information-sharing covers three key categories, restricted to FMF member firms:
Vulnerabilities, weaknesses, and exploitable flaws that may compromise the safety, security, or intended functionality of frontier AI models, such as jailbreaks, adversarial inputs, or data poisoning.
Threats directed at unauthorized access or manipulation of frontier AI models, including potential threat actors, attack vectors, and cyber-threat indicators.
Capabilities of Concern referring to frontier AI capabilities with potential for large-scale societal harm, such as CBRN development, offensive cybersecurity attacks, and model autonomy.
Executive Director Chris Meserole stated: "Responsible disclosure is core to our mission of advancing frontier AI safety and security." The organization aims to refine its information-sharing function over time while supporting collaboration with the broader frontier AI ecosystem.