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…

For the next ~8 hours I'll be in office working on a training I'm giving next week. To make things a little more fun I will be checking back here every so often to answer any questions you might have about offsec, red teaming, ai, vuln reserach, career stuff, etc. Dms open too

I think we've discovered another critical vulnerability: CVE-2026-BURNOUT: AppSec Professional Exhaustion Severity: Critical (CVSS 10.0) Description: A vulnerability affecting security professionals exposed to a continuous stream of critical RCEs, zero-days, emergency patches, and Friday afternoon disclosures. Impact: Sleep deprivation, excessive caffeine consumption, and an inability to enjoy weekends. Attack vector: Reading security advisories. Affected versions: All AppSec professionals, regardless of experience. Mitigation: No known patch. Air-gapping may delay the inevitable. Total system collapse remains a possibility. 🔥

🚨 Yesterday, PHP2Shell made headlines. Today, we're already teaching you how to exploit it. Yeah, we're THAT fast⚡ Two critical PHP vulnerabilities. 9.2 CVSS. Unauthenticated RCE. And you can already exploit the chain, bypass ASLR, and pop a shell in our latest recent threat room. Hackers don't wait. Neither do we 🟢 Go Get hands-on: tryhackme.com/room/cvephp2sh…