AI for Empowerment. We build private systems for enterprise, and open-source models for builders, to help people lead better lives.
Joined November 2020
- Tweets2.2K
- Following48
- Followers127K
- Likes4.5K
We're proud to share that Cohere's Chief AI Officer @joellepineau has been named to the @Forbes CIO Next list, which highlights the best and brightest C-level executives in the field of technology & security information.
Read her full profile here: forbes.com/sites/richardniev…
We’re looking for human-centred solutions, where AI enhances, not replaces.
Submissions will be reviewed by a panel of judges including @1vnzh. The top team receives a $3k prize, accelerated Cohere internship interviews, API credits & merch.
Register: sric.utoronto.ca/hacking-hig…
AI doomers' safety solution is always less competition, and @aidangomez wants the public to ask why.
Cohere believes technology should be something that you control, not that controls you.
Try Rerank 5 - the world’s smartest reranker - for more accurate search and RAG: openrouter.ai/cohere/rerank-…
2
15
2,316
When more people can access open models, everybody wins. We're adding as many open, world-class models as we can to Model Vault - both ours and others. Run them on dedicated or fully encrypted infrastructure.
With Model Vault, your models run on single tenant infrastructure that belongs to your workloads. No shared GPUs, no neighboring tenants, no infrastructure management.
For your most sensitive work, Model Vault Encrypted protects data in use, not just at rest and in transit. Nobody can see what you send to the model - not even us.
From growing our workforce to investing in the next generation of talent and research partnerships, we’re committed to building our future in Canada.
Read more: cohere.com/blog/cohere-canad…
We're humbled by the support from the country that raised our founders and defined our values. Looking forward to continue building in Canada, for Canadians 🇨🇦
"Cohere is a Canadian success story being written before our eyes" - Prime Minister @MarkJCarney
In the spirit of transparency and openness, we are releasing the traces so developers can test this out for themselves. We noticed this behaviour without prefilling with the following prompt, but this is rare. If you want to reproduce it, you can use this prompt with prefilling.
Open models make for better science and are part of why this problem came to light at all: developers running Qwen3.8 on their own machines spotted the strange links, and having the weights let us study the behaviour in depth.
We're encouraged by the thoughtful discussion our research has sparked across the AI and security communities. It's our belief that open-weight models are an important part of the future of AI, and we've recently released several under the Apache 2.0 license.
Trace, raw tokens, & minimal reproduction script: huggingface.co/datasets/Cohe…
While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected.
Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions.
Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent.
But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models?
We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem.
Read the full report: cohere.com/blog/concerning-b…
Trace, raw tokens, & minimal reproduction script found here on @huggingface: huggingface.co/datasets/Cohe…
Cohere's Model Vault is now available in public beta.
With @NVIDIAAI Confidential Computing, and GPUs on @Microsoft Azure, anybody can run our models while keeping their information secure.
Try it out today: cohere.com/blog/introducing-…
LLMs are getting better and better at math & science. But surprisingly, many can't solve a basic riddle.
@Cohere_Labs' CulturalRiddles is a new benchmark measuring culture as intertwined, lived experiences. True to open science, it was made by an unlikely crew.
The idea began with Hungarian diplomat Boczka Károly and quickly ballooned, with contributors such as high schooler Henry Gagnier and retired math teacher Jane Smith joining forces.
The team gathered collaborators to write original riddles about their own culture, in their own languages - and ended up with 6.2k of them, from 45 countries.
