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: https://t.co/TZSugXi2ZE
The leak of two critical PHP vulnerabilities has put millions of servers worldwide and thousands in South Africa in danger of remote execution attacks.
https://t.co/hbVOEKBibI
Please retweet. @Wise has blindsided us and put our business at risk. This is insane.
We’ve been a Wise customer for years, using their USD bank account to accept payments from our US clients.
Today, I discovered that Wise has capped how much we can send or withdraw from our account at roughly 40% of our usual monthly volume.
No advance notice. No explanation.
We’re suddenly at risk of being unable to make payments on time.
Luckily, we have other accounts. Some clients also pay us in Bitcoin. But most of our money comes into Wise.
How can @Wise do this after we’ve been a customer for years and processed millions of dollars through it?
Customer support can’t help us. We’ve filed a complaint, but who knows if it will go anywhere.
In the meantime, we can’t get our money out. We’re being held hostage.
"I lost my luggage on RwandAir. After its CEO told me they would not do anything, I came to Twitter to call them out. On my next trip to Rwanda, 8 men in suits came to my hotel and took me in and kept me in prison for 2 weeks."
Introducing The Cheeky Pint Benchmark.
We made our intern watch every episode of Cheeky Pint to bring you this breakdown of pints drank by guests to the nearest sixth.
Congratulations to @elonmusk who is currently in the lead with three full pints of Guinness.
I think the plan is to get researchers to start using their AI with a limited compute budget, then once they have all the academic data, they’ll basically steal all their ideas from underneath them and run the research internally with their unlimited compute budget!