Berryville Institute of Machine Learning
BIML is an independent nonprofit research institute focused on machine learning security, specifically the work of building security into ML systems at the design level.
BIML is an independent nonprofit research institute focused on machine learning security, specifically the work of building security into ML systems at the design level.
People
Updated 05/18/26Co-founder and CEO
Co-founder and researcher
Executive Director, Researcher, AI, MLsec, LLMs
Co-founder and researcher
Co-founder and research engineer
Funding Details
Updated 05/18/26- Annual Budget
- -
- Current Runway
- -
- Funding Goal
- -
- Funding Raised to Date
- $150,000
Org Details
Updated 05/18/26The Berryville Institute of Machine Learning (BIML) was founded in May 2019 in Berryville, Virginia, by Gary McGraw, Harold Figueroa, Victor Shepardson, and Richie Bonett. McGraw, who holds a dual PhD in Cognitive Science and Computer Science from Indiana University and spent 24 years as a software security expert and CTO at Cigital (acquired by Synopsys), formed BIML upon his retirement to apply the lessons of security engineering to the emerging field of machine learning. BIML operates as a small, distributed research group operating largely remotely, with its principal base in Clarke County, Virginia, at the foot of the Blue Ridge Mountains. The institute is organized as a 501(c)3 nonprofit corporation in Virginia and received IRS nonprofit status after a year-long application process. BIML's research focuses on machine learning security — the systematic identification of security risks built into ML systems themselves, not the use of ML as a security tool. The organization approaches this through architectural risk analysis (ARA), a methodology drawn from software security engineering. Its foundational 2020 publication introduced the BIML-78, a catalogue of 78 security risks applicable to all ML process models. In January 2024, BIML published a follow-on architectural risk analysis of large language models, identifying 81 LLM-specific risks and flagging 10 as most critical. Beyond these flagship reports, BIML maintains an annotated bibliography of ML security literature (updated through at least 2025), publishes articles in peer-reviewed venues such as IEEE Computer, hosts the Silver Bullet Security Podcast, and contributes to the ML security community through conference presentations, tutorials, and its interactive online risk framework. BIML received a $150,000 grant from Open Philanthropy in January 2021 to support research and outreach. The institute remains active as of early 2026, with blog posts and conference appearances continuing through March 2026.
Theory of Change
Updated 05/18/26BIML believes that many of the most serious risks from ML and AI systems arise from security vulnerabilities baked in during design and development, not just from deployment failures or misuse after the fact. By rigorously cataloguing these risks using proven security engineering methods (such as architectural risk analysis) and publishing accessible frameworks, BIML aims to equip ML developers and engineers with the knowledge to build more secure systems from the ground up. The causal chain is: identify and systematize ML security risks -> disseminate findings widely through publications, frameworks, and talks -> practitioners adopt security-aware design practices -> ML systems are built with fewer exploitable flaws -> reduced likelihood of harmful misuse, catastrophic failures, or unintended consequences at scale.
Grants Received
Updated 05/18/26Projects– no linked projects
Updated 05/18/26Discussion
No comments yet. Be the first to share your thoughts.