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:

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'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:

Octave (formerly @HexagonSIG) uses Cohere Embed on @awscloud Bedrock to power multilingual and multimodal search across their regulated enterprise data One of hundreds doing the same. Embed offers premium accuracy, lower vector storage costs, and is compatible with your data. The post:

Some strong embeddings models have been released recently. Ours is still better. We benchmarked Embed 5 Fast against Google's EmbeddingGemma 2 across 163 datasets using RCP-nDCG@10. Embed 5 Fast leads on over 95% of them.

If your company just tightened the token budget, they should probably try out North Admin. Control your models by the who, what, when, and where, and track and manage tokens in real time.

We are hiring for our growing RL Environments team at Cohere. We build replicas of real-world enterprise setups across many industries, put our AI agents in them and watch the agents train and improve. If you know anyone excited to work at the intersection of RL x Data x Engineering, send them our way! Or if that's you, join us:

Last week, we dropped North 2, our biggest upgrade to North yet. Here's what's new, including: - Admin features like cost controls and analytics - Cleaner, sleeker agent harness - Connectors to all your favorite apps Swipe through to see more ➡️

From Berlin to British Columbia, we're building a global company that lets you appreciate more of the globe - technology that empowers you to stay away from your work laptop. There's a lot of beauty in the world. Go out and see it. Cohere handles the rest.

Satya is spot on. The interesting engineering problem isn’t just making models behave, it’s building systems that don’t need to trust them in the first place. We’re working on some pretty fun problems @cohere. From detecting intent drift across long-running agent sessions, correlating eBPF runtime telemetry with agent actions, capability-scoped identities and enforcing security boundaries below the harness. Lots of interesting work at the intersection of systems engineering, security, and AI. We’re hiring engineers who want to build this stuff :)

Paris, let's chat. Come join us for a tech meetup next week, where we'll be discussing our research & engineering! Hear from our director of training integration, and speak with recruiters about life at Cohere. See you there 🇫🇷 Sign up here:

If you're a Canadian in the Bay Area and missing home, @brhydon knows how you feel. Join us at the Cohere office in SF next week to celebrate Thanksgiving, and learn about Cohere & what we're building up north:

Cohere is hosting an X Live this Thursday to discuss all things search & retrieval, including our recent model releases, latest research output, and what's planned for the rest of the year. It's a chance for our community to hear first-hand from Cohere's engineers and ask questions. If it goes well, we'll do more. Make sure to add it to your calendar:

At the beginning of 2026, we had ~400 employees. Now, we're over 800. Hear from Ana Cismaru, one of our senior product managers, about why she joined Cohere and how her projects & the people keep her motivated. We're still growing. Come join us today:

When you run your own business, the last thing you need is one more task on your plate. Focus on the thing that needs your attention most: your creativity. We can take care of the rest. @SRNA_LI empowers herself, and her business with Cohere.