AI Safety Fund Dec 2025 round

The Frontier Model Forum's AI Safety Fund (AISF) is a multi-lab consortium fund jointly underwritten by frontier model developers, distinct from any single lab's internal grant program. The December 2025 round disbursed more than $5M across 11 grantees with thematic scope spanning scheming detection, biosecurity, decentralized AI oversight, cybersecurity evaluations, and multi-agent safety.

The 11 named grantees and their projects are: Apollo Research ("Building black box scheming monitors for Frontier AI agents"); California Institute of Technology ("AI-driven Detection of Protein Mimetic Biothreats with BioSentinel"); Institute for Decentralized AI, operated by the Cosmos Institute ("Scalable, Decentralized Oversight for Multi-Agent Networks"); Faculty AI ("Automated Red-Teaming for Biosecurity Risks"); FAR.AI ("Quantifying the Safety-Adversary Gap in Large Language Models"); FutureHouse, Inc. ("Pioneering AI-Driven Experimental Design: Benchmarks for Responsible Innovation"); Morgan State University (comparative analysis of human cybersecurity performance with and without AI tools); Nemesys Insights LLC ("ICS Benchmark and Human Uplift Study"); SecureBio (evaluations of agent AIs executing large-scale-harm tasks); University of Illinois Urbana-Champaign (cybersecurity risk evaluations of AI agents with computer interaction); and University of Toronto ("Analyzing the Emergent Role of Sanctioning in Regulating Multi-Agent LLM Systems").

The biosecurity sub-cohort (Caltech BioSentinel, Faculty AI, SecureBio, plus the Apollo and Nemesys cyber work) tracks the AI-bio risk thread that AISF has emphasized across rounds. Apollo Research's black-box scheming monitor work and FAR.AI's safety-adversary gap project address the deception-and-control problem from complementary angles: behavioral monitoring versus adversarial measurement. The Institute for Decentralized AI grant is the structural outlier, funding oversight frameworks for networks of agents that do not run inside a single trust boundary.

AISF routes industry capital from Anthropic, Google, Microsoft, and OpenAI (the original Frontier Model Forum members) through a neutral consortium toward external researchers, with funding decisions and grant administration handled at consortium level rather than by any individual member. Compared to single-funder safety RFPs of similar size, the structure is a useful precedent for how frontier-lab money can reach the wider safety field without routing through one lab's brand.

The fund's earlier rounds focused heavily on biosecurity; the December 2025 round broadened into scheming detection and multi-agent governance, suggesting a deliberate widening of scope as more empirical work in those areas becomes tractable.