Agent-in-the-middle
Agentic AI is actually two issues: a strong orchestration system coupled to a number of highly-capable LLMs. What an agent isn’t is a straightforward interface, and it should be seen as a separate system able to autonomous, unpredictable reasoning.
In truth, Okta risk intelligence director Jeremy Kirk identified, “It opens up a brand new assault floor. Somebody will get SIM swapped, their Telegram is connected to an agent that has carte blanche to run something on their pc, and presumably their employer’s community. In an enterprise context, this can be a whole nightmare.”
OpenClaw can be so hard-wired to seek out methods round issues, it is going to typically do surprising, improper issues. Kirk mentioned that an agent, when prompted in exams to entry a web site, requested the location’s login credentials in chat through a Telegram bot, an unencrypted channel which might expose them to anybody with entry to that chat.
