Entity Profile: Machine Intelligence Research Institute (MIRI)
Overview
The Machine Intelligence Research Institute (MIRI) is a non-profit research organization based in Berkeley, California, dedicated to identifying and managing the existential risks associated with the development of smarter-than-human artificial intelligence (AI). Founded in 2000 (originally as the Singularity Institute for Artificial Intelligence), MIRI was one of the first institutions to frame AI safety as a technical mathematical problem.
MIRI has undergone a significant strategic pivot. After more than two decades of focusing primarily on technical "alignment" research, designing mathematical architectures to ensure AI systems remain beneficial?the organization, has shifted its priority toward Technical Governance and Public Policy Communications. This shift stems from their assessment that technical safety research is not moving fast enough to keep pace with the current rapid advancements in frontier AI models.
Continue…
Core Operations and Functions
MIRI functions as a specialized think tank and research house that operates at the intersection of mathematical logic, computer science, and international policy. Its activities are currently focused on three primary areas:
1. Technical Governance Research
MIRI?s "Technical Governance Team" (TGT) conducts research into the practical, technical mechanisms required to enforce AI safety at a global scale. Unlike traditional policy work, this research focuses on the "how" of enforcement, such as:
* Verification Mechanisms: Developing technical protocols to verify that nations or corporations are complying with international AI development limits without requiring intrusive constant surveillance.
* Compute Governance: Analyzing the hardware supply chain (specifically GPUs and AI chips) to identify "choke points" that could be used to regulate the training of dangerously large models.
* Legal and Constitutional Analysis: Investigating the intersection of AI regulation with existing laws, such as First Amendment constraints on model licensing and the executive powers required to halt hazardous development.
2. High-Stakes Public Communications
Recognizing that the "alignment problem" may not be solved before the advent of artificial superintelligence (ASI), MIRI has expanded its communications arm to engage with policymakers and the general public.
* Advocacy for a "Pause": MIRI is a leading voice advocating for a global, indefinite halt on the development of AI models more powerful than current frontier systems (such as GPT-4) until safety can be mathematically proven.
* Briefings and Explainers: The institute produces distilled technical briefings for governments, such as their recent 2025/2026 reports on why extinction is the "default" outcome of unaligned superintelligence.
3. Foundational Alignment Research (Maintenance)
While it is no longer their primary focus, MIRI maintains a repository of foundational research that has influenced the broader AI safety field. This work centers on "Agent Foundations"?the mathematical study of how rational agents should behave.
Products and Technical Offerings
MIRI?s "products" are primarily intellectual?consisting of peer-reviewed papers, technical reports, and open-source software tools.
1. Technical Reports and Papers
MIRI produces high-impact documentation used by researchers and policymakers. Key categories include:
* Strategic Landscape Papers: Reports like AI Governance to Avoid Extinction that map out actionable research questions for global security.
* Technical Proposals: Formal drafts, such as the International Agreement to Prevent the Premature Creation of ASI, which provide a ready-to-use framework for international treaties.
2. Software and Algorithmic Tools
Over its history, MIRI has released or influenced several technical tools:
* Logical Induction: A computable algorithm that assigns probabilities to logical statements and refines them over time, addressing how AI might reason about its own code.
* Corrigibility Frameworks: Mathematical models for "utility functions" that incentivize an AI to allow itself to be shut down.
3. Educational and Training Programs
- MIRI Research Fellowships: In early 2026, MIRI launched an 8-week intensive fellowship program in Berkeley to train a new cohort of researchers in technical governance.
- Research Workshops: The institute hosts invitational workshops where mathematicians and computer scientists collaborate on foundational safety problems.
Strategic Services
- Policymaker Advising: MIRI leadership and senior researchers provide direct consultations to government bodies on the catastrophic risks of AI and the technical feasibility of different regulatory approaches.
- Grantmaking and Support: Through the "MIRIx" program and other initiatives, the institute has historically provided funding and support for independent research groups working on alignment-adjacent problems.
Impact and Philosophy
MIRI is characterized by its "pessimistic" but rigorous outlook on AI safety. Their work assumes that creating a superintelligent agent that does not destroy its environment is an extraordinarily difficult task, one that requires solving deep problems in decision theory and mathematical logic. By shifting toward governance, MIRI acts as a "safety brake" for the industry, providing the technical and strategic arguments for why the world must slow down its pursuit of general artificial intelligence.
