Last successful capture October 13, 2026 · text unchanged since October 8, 2026
Get the research letter
Companies change their terms quietly. We read every version and catch what actually changed. One email a week on the changes that matter and what they mean.
This page describes what the document states, permits, or reserves. It does not constitute a legal determination about enforceability. Regulatory applicability may vary by jurisdiction. Methodology
13Total
8High severity
4Medium severity
1Low severity
Stay ahead of the changes
Track Cohere and get the diff the day its terms change.
This policy sets out what Cohere forbids you from doing with its AI models: you cannot use generated text to fake public opinion, make important decisions about people without a human involved, spread misinformation, or produce hate-inciting content. Cohere also warns that its Command R models should not be used alone for decisions that significantly affect individuals, such as in financial services, employment, or housing. Cohere can restrict API access at any time if it determines your use case violates its guidelines.
Analysis
Cohere's Responsible Use Policy establishes a set of high-severity prohibitions and use-case restrictions governing how users and developers may deploy Cohere's language models, including generated text and API access. The policy prohibits astroturfing, human-outside-the-loop decision-making, generation of misinformation, opinion-manipulating content, and hate-inciting or group-mischaracterizing material. It restricts use of Command R models as standalone tools for high-stakes individual decisions in domains such as financial services, employment, and housing, and states that language models should not be used to make decisions about individuals or the groups they belong to. Cohere reserves the right to restrict API access at any time for malicious use cases that violate its Usage Guidelines and Terms of Use, and directs users and developers to those separate governing documents for the full enumeration of prohibited use cases.
What this means for you
For an individual using Cohere's platform, this policy means that certain high-risk uses are prohibited outright — including generating content designed to simulate public opinion, producing news or articles intended to manipulate opinion, and using the model's outputs to make important decisions about people without human review. Cohere acknowledges that even with safeguards in place, harmful or toxic outputs can still occur, particularly in extended multi-turn conversations. Users can consult Cohere's Usage Guidelines page, which Cohere explicitly recommends as a resource for understanding how to use the platform responsibly.
Institutional Analysis
Stay ahead of the changes
Institutional analysis available with Insight
Which mapped governance frameworks each document engages, tied to the specific provisions that engage them.
Our reading · This prohibition covers a broad range of influence operations, explicitly extending the ban beyond social media to any channel where artificial public opinion could be manufactured.
In ConductAtlas since July 10, 2026StandardSeen across 274 platforms
Our reading · This reservation means API access is not guaranteed to continue, and developers or users building on the API may lose access without a specified prior notice or cause requirement.
In ConductAtlas since July 10, 2026StandardSeen across 251 platforms
Our reading · This non-recommendation identifies high-stakes individual decision-making as a domain where the Command R models alone are considered by Cohere to be unsuitable.
In ConductAtlas since July 10, 2026StandardSeen across 238 platforms
Our reading · This prohibition requires human oversight in any process where Cohere-generated text informs important decisions about individuals, limiting fully automated decision-making pipelines.
In ConductAtlas since July 10, 2026StandardSeen across 274 platforms
Our reading · This restriction extends beyond individuals to group-level decisions, broadening the scope of prohibited uses tied to known model biases.
In ConductAtlas since July 10, 2026StandardSeen across 238 platforms
Our reading · Cohere links the API access restriction right explicitly to malicious use cases that breach its Usage Guidelines and Terms of Use, establishing the policy basis for enforcement.
In ConductAtlas since July 10, 2026StandardSeen across 274 platforms
Our reading · This prohibition covers two distinct categories of harmful content — opinion manipulation through fabricated or misleading articles and hate-inciting or group-mischaracterizing content — establishing broad limits on permissible generation.
In ConductAtlas since July 10, 2026StandardSeen across 238 platforms
Our reading · The clause identifies two separate governing documents — Usage Guidelines and customer agreements — as the locations where prohibited use case details are set out, and names three specific prohibited categories.
In ConductAtlas since July 10, 2026StandardSeen across 274 platforms
Our reading · This recommendation identifies two specific risk categories that Cohere singles out as deserving heightened developer attention when building on its platform.
In ConductAtlas since July 10, 2026StandardSeen across 268 platforms
Our reading · The reliability limitation on non-English outputs may affect the suitability of Cohere's model for multilingual or non-English use cases.
Our reading · Cohere's acknowledgment that toxicity remains possible despite safeguards is consequential for users and developers who may encounter harmful outputs during extended interactions.
In ConductAtlas since July 10, 2026CommonSeen across 139 platforms
Complete Provision Index
Every distinct legal provision identified in this document. Featured provisions appear above with analysis.
We have not recorded a change to this document since we began capturing it. Monitor includes same-day alerts, structured change summaries, and monitoring for up to 20 platforms.