As a leading provider of enterprise AI solutions, Cohere delivers the technology foundation that powers countless enterprise AI applications and products. And that foundation needs to be trustworthy. Trust, built upon security, safety, transparency, reliability, accountability, and state of the art research, enables our customers to innovate with confidence and accelerate their AI deployments.

Overview

Cohere’s approach to AI governance includes frameworks, practices, and policies that direct and guide the responsible development and use of our AI models and systems. This includes providing our customers with the information they need to deploy and build high trust enterprise AI solutions on top of those models and systems.

Cohere’s AI governance approach is aligned with and certified under ISO/IEC 42001, which ensures systematic and ongoing risk management of AI systems, robust governance practices that are regularly reviewed, and external validation steps.

Our approach is also driven by international voluntary standards for AI developers, which provide principles-based frameworks that foster harmonization and trust across the AI industry. Since 2023, Cohere has signed onto several voluntary commitments and frameworks for AI governance, including:

The Seoul Frontier AI Safety Commitments and the Seoul AI Business Pledge
Canada’s Voluntary Code of Conduct on the Responsible Development and Management of Advanced Generative AI Systems
The EU AI Act’s voluntary General-Purpose AI Code of Practice for general-purpose AI models and AI Pact

We also contribute to and align our approach with other leading frameworks for AI governance, including America’s AI Action Plan and the Hiroshima Process International Code of Conduct for Organizations Developing Advanced AI Systems.

Cohere’s AI governance practices

Our governance practices are organized around the following core pillars of trust: Transparency and Explainability; Security and Data Protection; Safety; Enterprise Performance, Reliability, and Efficiency; Accountability; and Advancing the State of the Art.

Transparency and Explainability

Cohere develops, publishes, and maintains a range of documentation that provides information to our customers, the research community, and the wider public about how we build our AI products and how they can be deployed responsibly. We also build features into our AI solutions that ensure AI outputs can be explained and understood.

Our transparency and explainability publications and practices include:

Our Secure AI Frontier Model Framework outlines our practices and protocols for identifying, testing, and mitigating risks throughout the model development lifecycle. This framework is aligned with multiple industry commitments, including the 2024 Seoul Frontier AI Safety Commitments.
Model cards and technical reports, and many other resources on Cohere’s documentation website, describe our models' capabilities, evaluation results, configurable safety features, limitations, and more. This enables our customers and other external stakeholders to understand the design and characteristics of our models, supporting transparency, explainability, and responsible use.
Policies outline the acceptable use of our models.
Built-in features in our AI models and North, our end-to-end AI platform, enhance transparency and explainability. This includes citing the sources on which specific outputs are based, and providing users with a description of the steps followed by an AI agent to generate an output in response to a user’s input. These features empower users to verify how the AI system generated a particular output and what steps and user-connected systems or sources were used to get there.

Security and Data Protection

As the leading security-first AI model developer, Cohere advances best practices in AI security and data protection in the enterprise space. This includes adhering to a range of industry standards and certifications for trust, security, and data protection that are important to our customers, such as:

ISO 42001: Developed and certified by the International Standards Organization, ISO/IEC 42001 is the world’s first AI management system standard. It requires organizations to demonstrate rigorous governance and transparency standards to effectively identify and manage the potential risks of AI systems.
ISO 27001: This is an internationally recognized standard for information security, showing that organizations have implemented a systematic approach to managing information security risk and protecting private information. Both ISO 42001 and ISO 27001 require regular surveillance and recertification audits.
SOC 2 Type 2 Report: This auditing procedure was developed by the American Institute of Certified Public Accountants. It assesses the effectiveness of an organization’s controls to manage customer data and ensure its confidentiality, integrity, and availability. We undertake an annual audit and make a report available to customers.
UK Cyber Essentials: This UK government-backed certification scheme is designed to help organizations assess and mitigate risks from common cyber security threats to their IT systems and demonstrate an appropriate level of protection.
Defence-in-depth: We take a defence-in-depth approach that includes red teaming and penetration testing. Details of our security testing and approach are available in our Secure AI Frontier Model Framework.

Safety

Cohere focuses on known and observable risks of harm to individuals, organizations, or society. We develop mitigations within Cohere’s models and products that enable enterprises to customize safety controls according to their business, domain, and country context. Our approach to safety is based on three core principles:

Evidenced Risks: We focus on real-world risks that are known, measured, or observable
Assessed in Context: We identify and assess risks in the context of Cohere’s and and our customers’ specific use cases
Holistically Managed: AI risk management is embedded by design into model and system development

Evaluations are conducted throughout the model development cycle, using both internal and external evaluation benchmarks. For example, we evaluate new model versions against industry-standard benchmarks like BOLD (Biases in Open-ended Language Generation) or custom internal benchmarks, and publish the results in technical reports. We also make our models available for evaluation to reliable, independent partners, such as the Cloud Security Alliance (CSA) Risk Rubric and the MLCommons AILuminate benchmark.

Enterprise Performance, Reliability, and Efficiency

From design to deployment, Cohere prioritizes enterprise performance, reliability, and efficiency, so that AI can be adopted in real-word applications. This includes:

Enterprise performance by design: Cohere’s enterprise focus drives our development process. Training data, model architecture, capabilities, and modalities are curated to optimize model performance in capabilities such as code, math, reasoning, multilingual, and safety, as well as information retrieval and tool use.
Multilingual capability: We ensure that all our models are performant across multiple languages, which include 23 major languages in the global enterprise space, such as French, Arabic, and Spanish. Multilingual performance is not only useful, it is critical to safe AI in real-world applications.
Testing and validation for the real world: As detailed in our technical reports, we rigorously test our AI models to assess enterprise capabilities and test performance in use cases that mirror enterprise applications. For example, benchmarks were developed and used to test both the correctness and accuracy of Command A outputs. Correctness tests assessed whether the output was correct based on the facts and context, while accuracy metrics assessed whether the model accurately identified whether or not a question was answerable based on the source documents provided to respond to a user query. Command A’s strong performance on the benchmark demonstrates robust mitigation of the risk of inaccurate or hallucinated outputs.
Inference efficiency: We prioritize models and AI solutions that can be served with minimal compute. This ensures that private and secure deployments are accessible to enterprises and can be deployed in the real world.

Accountability

At Cohere, we believe that AI governance requires multidisciplinary collaboration, that decisions must be owned by product and modelling teams responsible for developing the technology, and that governance responsibilities are shared along the AI value chain.

We ensure multidisciplinary collaboration and accountability in our own governance program through:

A cross-functional leadership committee that provides internal oversight and strategic direction for Cohere’s Artificial Intelligence Management System (AIMS), ensuring executive support and review of Cohere’s AI governance practices.
A cross-functional team of technical and non-technical staff from the Policy, Legal, Safety, Data and Modeling, Research, and Security teams. This group is responsible for operationalizing governance mechanisms in their respective functions, such as developing policies, conducting security audits, carrying out safety testing, sharing insights and developments in AI governance, and advancing AI governance objectives at Cohere.
Clear, final points of evaluation to validate that a model is safe and secure, and that accountability for decisions to make a product available to our customers is escalated to our C-suite leaders.

We are accountable to our customers through our contractual commitments and assurances, including a copyright indemnity. We also provide clear guidance to customers, through our documentation and our direct engagements, about their own governance responsibilities for the solutions they develop using Cohere models and products.

Advancing the State of the Art

Cohere's research organization, Cohere Labs, supports fundamental research to advance the state of the art, promote safe and transparent AI development, and broaden global participation in AI research. Our mission is to both expand human knowledge and increase equitable access to the tools that make such research possible. These efforts include:

Fostering knowledge sharing among the open science community by collaborating with the global research community and supporting reproducible research practices
Developing AI talent around the world by providing access and financial support for AI research using Cohere models
Conducting open research to advance and inform AI governance, including multilingual capabilities and AI safety, model efficiency, compute thresholds, and AI risks and harms
Releasing open weights models to enable transparent, peer-reviewable research on model behavior, performance, and safety

We believe that responsibly releasing model weights is critical to building a trustworthy AI ecosystem. It supports scientific rigor by enabling reproducibility, democratizes innovation, and allows external researchers to audit, stress-test, and improve models. This openness strengthens collective understanding of AI models, improves robustness, fosters better safety practices, and promotes accountability across the ecosystem.

Global engagement and cooperation for a trustworthy AI ecosystem

AI governance plays a critical role in building and maintaining trust in AI technologies. Cohere is committed to engaging and cooperating with the broader AI ecosystem to enhance trust through:

Standardization: We collaborate with leading safety and security research institutes, including the US National Institute for Standards and Technology, the Canadian AI Safety Institute, and the UK AI Security Institute, as well as with leading industry working groups to develop AI safety and security standards, such as MLCommons, the OWASP Generative AI Security Project, and the Coalition for Secure AI
Consultation: We actively contribute to and participate in international governance development processes, such as the process to develop the EU AI Act’s General-Purpose AI Code of Practice for general-purpose AI models or the Government of Canada’s AI Advisory Council, to share our practical experience and expertise with policymakers
Education: We produce AI literacy materials for developers, businesses, and the general public, such as our LLM University courses and Enterprise Guide to AI Safety

Learn more

To explore Cohere’s full approach to secure AI, governance, and enterprise readiness, visit our Trust Center.

Building trust in AI: Cohere’s approach to AI governance