OpenAI has reported 2 more incidents of rogue AI agents, this time during third-party testing
THE SO WHAT
Third-party testers triggering “rogue” agent behavior is a reminder that misconfigurations, not just model intent, drive real-world risk. If you’re piloting agentic systems, treat environment setup, tool scopes, and kill switches as first-class engineering artifacts, not afterthoughts.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIHow Reddit moderators are battling AI-powered astroturfing ad campaigns that exploit the platform's trusted reputation to promote skincare and other products
AI-powered astroturfing on Reddit is eroding the last pockets of perceived “authentic” UGC. If your brand or product depends on subreddit trust, assume organic channels are now adversarial and invest in verification, mod relationships, and off-platform proof.
Applied AIWe're building the most critical infrastructure of our lifetimes. We're leaving the front door unlocked
AI data centers are being treated as software projects on hardware timelines—security is the variable line item. If you run or depend on large AI clusters, move physical security and insider threat from facilities to board-level risk this quarter.
Applied AI'Patients are ready for this': New study reveals 90% of NHS staff use AI at work — and most patients are happy with it
If 90% of NHS staff are already using AI—mostly for admin—and ~70% of patients feel positive, the adoption bottleneck is now governance, not appetite. Health operators should formalize shadow AI into sanctioned workflows before regulators and unions do it for them.
Applied AIZero-Mem: Zero-Token Memory Operations for LLM Agents
Zero-Mem’s “zero-token” memory operations aim to decouple agent memory cost from context length—if it works in practice, it changes the unit economics of long-horizon agents. Infra teams should track whether techniques like this make it into mainstream stacks before overprovisioning for context-heavy workloads.