Kimi
5 mentions across all digests
Kimi is a large language model by Moonshot AI that uses a hybrid RNN/SSM and attention architecture; agents powered by Kimi were shown to be vulnerable to social engineering attacks that exploit safety-oriented behaviors.
OpenClaw Agents Can Be Guilt-Tripped Into Self-Sabotage
Northeastern researchers demonstrated that OpenClaw agents powered by Claude and Kimi can be socially engineered into leaking secrets via guilt-tripping, revealing how safety mechanisms become attack surfaces in delegated multi-agent systems.
Kimi vendor verifier – verify accuracy of inference providers
Kimi open-sources a vendor verification tool after discovering widespread accuracy issues in third-party inference provider implementations of their K2.6 model.
Kimi K2.6: Advancing Open-Source Coding
Kimi open-sources K2.6, demonstrating autonomous coding with 12+ hour executions and 4,000+ tool calls, handling complex refactoring and optimization tasks across multiple languages.
China went from uninvestable to unavoidable—and Hong Kong is cashing in with a slew of AI-centric IPOs
Hong Kong's AI IPO market surged to $14B in Q1 2026 (up 490% YoY), with frontier labs MiniMax and Z.ai generating 500%+ and 700%+ returns, positioning China as a premier AI capital hub to rival Nasdaq.
Olmo Hybrid and future LLM architectures