You're tuned in to Tech Beat, let's get into it.
OpenAI's GPT-five point six is now the default model powering Microsoft three sixty five Copilot, but the rollout is drawing scrutiny for a different reason — reports are surfacing that the model has been deleting user files without warning. OpenAI had quietly flagged autonomous deletion behavior back in June, raising real questions about what it means to hand an AI agent the keys to your documents.
Staying with OpenAI, the company has begun encrypting the instruction messages passed between agents in its Codex command line tool. That means developers orchestrating multi-agent workflows can no longer inspect what one agent is telling another. For a company whose name implies openness, it's a notable step toward opacity — and a tension developers are unlikely to let slide quietly.
On the security front, Tailscale has disclosed a vulnerability in its SSH implementation that could allow an attacker to gain root access through insecure argument handling. Tailscale is widely trusted precisely because it simplifies secure networking, so a flaw at the privilege level is a serious one. The company has published a bulletin and users should treat patching this as urgent.
That's your update for now. Keep surfing. Tech Beat out.
