ResearchSplitAgent: Privacy-Preserving Distributed Architecture for Enterprise AgentsarXiv·high signalXBlueskyLinkedInCopy linkSplits agent execution between enterprise privacy agents and cloud capability agents. Addresses MCP/A2A assumption of complete data sharing.SourceSource pagearXiv↳ Follow the threadPolicy dependency / Stack layerCompose agent guardrails as an algebra instead of a rule list: 94.8% of policy-violating events intercepted while keeping 86.9% task completionarXiv 2608.16402Stack layer / Threat patternMCP tools that write, not read, went from 27% to 65% of tool use - and measured defenses stop under 30% of attacksarXiv 2608.17275Stack layer / Threat patternSkillWatermark: benign-looking skill descriptions turn agent network traffic into a covert exfiltration channelarXivPolicy dependency / ContrastZetta ζ Tops HuggingFace Daily Papers at 152 Upvotes: Robots That Update Skills Mid-Task, 90.8% on LIBERO-Pro and 11.1x Faster InferencearXiv (via HuggingFace Daily Papers)Stack layer / ContrastOmniScientist Runs Ideation, Experimentation and Manuscript Writing Across 36 Real Cases and Wins 85% of Head-to-Head Comparisons Against a Features-Only BaselinearXiv (via HuggingFace Daily Papers)Policy dependency / Stack layerBlack-box RL through an unmodified agent harness lifts Pass@1 by 9.98-14.81 points, with Claude Code as one of the harnessesarXiv 2608.16798Stack layer / Update threadTrain the harness, not just the model: 6K examples of harnessed agentic RL lifts Qwen3.5-9B on SWE-bench Verified from 41.8% to 56.4%arXiv 2608.17528Stack layer / ContrastBenchmarked against workflows real users already pay for, the best agent finishes only ~30% end to endarXiv 2608.17800