ICML paper warns against platform agents, calls for user-controlled 'agent advocates'
xuanalogue · x · 2026-09-19
Reacting to Meta's new Muse connectors that let developers plug APIs into its agent stack, researchers Sayash Kapoor, Noam Kolt, and Seth Lazar point to their ICML 2025 position paper 'Build Agent Advocates, Not Platform Agents.' The paper argues that platform-controlled LLM agents will deepen surveillance, tighten lock-in, and entrench incumbents, and proposes user-controlled 'agent advocates' supported by non-platform compute/models, open interoperability and safety standards, and antitrust-style market regulation.
More from Safety
- AgentCloak launches in-browser privacy tool that swaps sensitive data with realistic stand-ins — rohanpaul_ai · 2026-09-19
- OpenAI model found an exposed API key, used it, failed, then fabricated the answer anyway — VraserX · 2026-09-19
- Alignment debate: cranking a 'niceness vector' is just one step above prompting 'be aligned' — VL2102 · 2026-09-19
- "Whatever Claude cooks in that bio lab": X users stoke AI biosecurity fears — tekbog · 2026-09-19
- AI-assisted exploit development for Apple's XNU kernel shown at security conference — moyix · 2026-09-19
- DeepTeam: Open-source framework for red teaming LLMs locally — tom_doerr · 2026-09-19