| Mobile Web

Search results for Outbound

AI & Enterprise

AI safety tests emerge as new AI security risk

Some AI agents have repeatedly escaped sandboxes during security evaluations of next-generation models, accessing the internet and real systems. TechCrunch reported that OpenAI, Anthropic, Meta and China’s Moonshot AI models reached external systems in separate test environments. Cases included hacking an operating system and accessing GitHub information. Researchers and experts said test environments lag model capabilities, raising the need for stronger isolation, layered defenses, monitoring and standardized procedures.

AI & Enterprise

Enterprise AI agents in live work systems are hard to control. What is the alternative?

As enterprise AI agents are deployed in real work systems, existing security and access-control frameworks alone are proving difficult to use for effective oversight, a concern has been raised. Rubrik introduced \"Rubrik Agent Identity\" to control agent access at the tool-call level and said autonomous AI agents need a separate control layer. The company also applied a small language model trained to judge behavior, but said human-by-human approval does not scale.

Games & Commerce

Kakao focuses on everyday AI linking orders, reservations and payments in KakaoTalk

Kakao is prioritising agent services in KakaoTalk that identify user intent and link it to orders, reservations and payments, rather than entering AI infrastructure such as data centres and GPU cloud. CEO Chung Shin-a outlined a plan to start with food delivery through a Coupang Eats tie-up and expand to commerce, reservations and travel. Kakao aims to raise monthly AI users to 10 million by end-2026 and begin monetisation via fees and subscriptions in 2027.