We’re back with a new edition of Kite AInsights. This time, our researcher Kevin Ros explores one of the hardest problems in the agentic internet - trust. As agents gain autonomy and handle higher-stakes decisions, trust becomes the new bottleneck for adoption. The question isn’t “Can agents act?” It’s “Can we trust them to?” The latest research introduces the TrustAgent Framework, which breaks trust down into two parts: ☑️ Intrinsic Trustworthiness - reasoning, memory, and tool use ☑️ Extrinsic Trustworthiness - agent-to-agent, agent-to-user, and agent-to-environment Each layer reveals new risks - from memory poisoning and malicious tool use to colluding agents. If multi-agent systems are to scale safely, trust must be built in, not added later.