AI Safety Companies

Roblox AI

San Mateo, United States

Roblox AI develops and deploys artificial intelligence solutions to power the Roblox platform, a user-generated massive multiplayer online game. Their core technology focuses on generative AI tools for content creation alongside AI-driven systems for content moderation and platform safety. Targeting the vast Roblox developer community and its millions of daily users, Roblox AI aims to scale content creation and maintain a safe, engaging virtual environment through automated solutions.

OpenAI

San Francisco, United States

OpenAI is an AI research and deployment company dedicated to ensuring that artificial general intelligence benefits all of humanity. Creator of GPT-4, ChatGPT, DALL-E, and Sora.

Anthropic

San Francisco, United States

Anthropic is an AI safety company working to build reliable, interpretable, and steerable AI systems. Creator of Claude, focused on Constitutional AI and harmlessness research.

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, builds more understandable and customizable AI systems. Raised a record $2B seed round.

Safe Superintelligence (SSI), founded by OpenAI co-founder Ilya Sutskever, focuses exclusively on building safe superintelligent AI.

OneTrust

Atlanta, United States

OneTrust provides a unified platform for privacy, security, and governance automation, enabling organizations to manage data and comply with evolving regulations. Their core offering centers on AI governance solutions that embed compliance controls throughout the entire AI lifecycle, from data collection to deployment. OneTrust targets enterprises seeking to operationalize responsible AI practices and mitigate risk while accelerating innovation, particularly those facing complex regulatory landscapes.

Applied Intuition

Mountain View, United States

Applied Intuition provides a comprehensive software platform, Vehicle OS, for developing and validating autonomous vehicle (AV) and advanced driver-assistance systems (ADAS). Their technology focuses on large-scale data ingestion, synthetic scenario generation, and machine learning infrastructure to enable rapid development, testing, and deployment of AI-powered vehicle intelligence. They primarily serve automotive manufacturers and technology companies seeking to accelerate AV/ADAS development and improve safety through data-driven insights and efficient software updates.

Kodiak Robotics

Mountain View, United States

Kodiak Robotics develops autonomous driving technology for long-haul freight, focusing on the deployment of self-driving capabilities onto existing truck platforms. Their core product, the “Kodiak Driver,” is an integrated AI-powered system combining hardware and software for reliable autonomous operation in diverse environments. Kodiak targets both commercial freight carriers seeking increased efficiency and the national security sector, offering a scalable solution to address driver shortages and improve operational safety.

Nauto

Palo Alto, United States

Nauto provides a predictive AI-powered driver and fleet safety platform for commercial vehicle operators. Their core technology utilizes in-cabin camera systems and real-time analysis to detect and alert drivers to risky behaviors – such as distraction or fatigue – before collisions occur, functioning independently of external connectivity. Nauto targets the commercial fleet market, offering a privacy-focused solution that reduces incidents, lowers insurance costs, and streamlines driver coaching through event-based video analysis.

Magic

San Francisco, United States

Magic is an AI company developing frontier-scale code models designed to automate software engineering and AI research. Their core technology focuses on ultra-long context language models, leveraging 8,000 H100s to improve model performance and address AI alignment challenges. Targeting the advanced AI research community and software development organizations, Magic aims to accelerate progress towards safe Artificial General Intelligence (AGI) through automated code generation and model improvement.

WhyLabs

Seattle, United States

WhyLabs was a US-based company that provided AI observability and data monitoring solutions for machine learning systems. Their core offering focused on detecting and preventing issues related to data quality, model performance, and responsible AI practices. While the company has ceased operations, WhyLabs contributed to the field through open-source initiatives and aimed to serve organizations prioritizing the reliability and ethical deployment of AI.

Cleanlab

San Francisco, United States

Cleanlab provides a post-hoc AI safety layer that identifies and remediates incorrect or unsafe outputs from any deployed AI agent, including those powered by large language models. Their core technology utilizes a proprietary confidence-based approach to detect label errors and predict potentially harmful responses without requiring retraining or modifications to existing AI infrastructure. Cleanlab targets enterprises prioritizing AI safety, compliance, and trustworthiness, offering a deployable solution for maintaining quality control over generative AI and other AI-driven applications.

Encord

London, United Kingdom

Encord provides data annotation and model evaluation tools for building safer AI, with focus on quality control and active learning.

Unitary

London, United Kingdom

Unitary delivers AI-powered automation for customer and marketplace operations through its Virtual Agent technology, which directly logs into existing tools and follows pre-defined processes without requiring API integrations. These Virtual Agents combine computer vision and complex reasoning capabilities to handle tasks like content moderation and policy enforcement, escalating to human experts when necessary to guarantee accuracy. Unitary focuses on rapidly deploying automation at scale for businesses seeking to reduce operational costs and improve customer satisfaction, offering a results-based pricing model with no upfront fees.

Conjecture

London, United Kingdom

Conjecture is a UK-based AI research company developing a novel AI architecture focused on verifiable safety and control. Their core technology centers on building AI systems with inherent transparency and predictability, moving beyond traditional black-box approaches. This positions Conjecture to serve organizations and researchers prioritizing robust AI safety and alignment, particularly as advanced AI capabilities continue to develop.

Saidot

Helsinki, Finland

Saidot is a Finnish SaaS provider specializing in AI governance solutions. Their platform enables organizations to document, assess, and monitor AI systems to ensure responsible development and adherence to the forthcoming EU AI Act. Saidot targets businesses and public sector entities requiring demonstrable AI compliance and risk management capabilities within the European regulatory landscape.

Mila

Montreal, Canada

and aiming for a professional, informative tone: Mila, founded in 1993 by Yoshua Bengio, is a leading Quebec-based research institute specializing in deep learning and generative models, with a particular focus on AI safety and responsible development. The institute coordinates the work of over 140 affiliated professors across multiple universities – including Université de Montréal, McGill, and Polytechnique Montréal – conducting research in areas like natural language processing and machine learning theory. Mila actively translates research into real-world impact through initiatives like the AI Policy Frontline and the TRAIL program, aiming to address critical challenges at the intersection of AI and public policy.

Redwood Research

Berkeley, United States

Redwood Research is a US-based AI safety nonprofit focused on mitigating catastrophic risks from advanced AI systems. Their core research centers on “AI control” protocols – techniques for reliably monitoring and preventing subversion by potentially deceptive large language models, even when those models intentionally conceal misaligned intentions. They serve as a critical resource for both governments and leading AI developers like Google DeepMind and Anthropic, providing expertise and methodologies for assessing and mitigating AI safety risks.

Apollo Research

London, United Kingdom

Apollo Research is a UK-based AI safety and alignment company specializing in the detection of deceptive behavior in advanced AI systems, particularly large language model agents. They develop and implement novel AI model evaluations focused on “scheming” – covertly pursuing misaligned objectives – and provide technical expertise to governments and international organizations on AI governance and regulation. Their core offering is third-party evaluation of frontier AI models, alongside consultancy services for responsible AI development frameworks and policy guidance.

Anthropic Research

Berkeley, United States

Anthropic Research develops robust evaluation frameworks, most notably ARC Evals, to rigorously assess the safety and reliability of large language models (LLMs). Their key innovation lies in utilizing model-written evaluations – leveraging LLMs themselves to generate challenging test cases and identify potential harmful behaviors in other models. This approach, coupled with interactive data visualization tools, allows Anthropic to systematically explore LLM behaviors and provide insights for improving alignment and reducing risks associated with frontier AI systems, serving researchers and developers focused on responsible AI development.

Center for AI Safety

San Francisco, United States

The Center for AI Safety is a US-based nonprofit focused on mitigating potentially catastrophic risks from advanced artificial intelligence. They conduct and fund research into AI safety, with a particular emphasis on identifying and addressing vulnerabilities in increasingly capable AI systems – offering resources like a dedicated compute cluster to support this work. Their primary audience includes AI researchers, policymakers, and stakeholders concerned with the responsible development and deployment of powerful AI technologies.

Optimus Ride

Boston, United States

Optimus Ride develops Level 4 autonomous vehicle technology specifically for low-speed, geo-fenced environments. Their core product is a full-stack autonomous shuttle solution, integrating hardware and software for first/last mile transportation. The company targets business campuses, residential communities, and tourist destinations seeking to deploy safe and efficient, on-demand mobility services within defined operational design domains.

Hyperloop TT

Los Angeles, United States

HyperloopTT is developing a full-scale hyperloop transportation system designed to move passengers and freight at airplane speeds. Their core technology centers on magnetic levitation within a near-vacuum tube, optimized by AI-powered control and safety systems. Targeting both passenger transit and express freight markets, HyperloopTT is currently focused on feasibility studies and pilot projects in regions like Brazil and Italy, aiming to deliver a sustainable and high-speed alternative to traditional transportation.

Anthropic Safety

San Francisco, United States

Anthropic develops advanced large language models, most notably the Claude family of AI assistants, with a core focus on safety and interpretability. Their key innovations include “Constitutional AI” – a technique for aligning AI behavior with a set of principles – and “circuit tracing,” which allows researchers to visualize and understand the internal reasoning processes within Claude models. Anthropic’s research and technology are targeted towards developers and enterprises seeking reliable and steerable AI, demonstrated by projects like “Project Vend” exploring real-world AI task completion and ongoing efforts to enable AI introspection and cross-lingual reasoning.

MIRI

Berkeley, United States

The Machine Intelligence Research Institute (MIRI) is a research nonprofit studying the mathematical underpinnings of intelligent behavior to ensure AI systems are safe and beneficial.