Microsoft interns studied making computer-use agents robust to misaligned incentives
ZacharyHuang12 · x · 2026-09-30
CMU student YuxuanL wrapped up a research internship at Microsoft AI Frontiers (now Microsoft AI), working with Zachary Huang and others on whether computer-use agents stay robust in incentive-misaligned environments and on training them to be more robust. A paper post is coming soon.
More from coding & agent
- OpenAI opens MCP Extensions to build plugins that feel like native ChatGPT features — haltakov · 2026-09-30
- Codex returns to the cloud just over a year after its first version — kevinkern · 2026-09-30
- Turso 0.8.0 cuts tail latency up to 500x vs SQLite under concurrent writes — solyarisoftware · 2026-09-30
- Vibe coding in 2026: 70% of project time still goes to edge cases humans must solve — kmeanskaran · 2026-09-30
- RuneScape Bench Turns an Open-Source MMO Into a Multi-Agent Testbed for LLM Agents — mrdrozdov · 2026-09-30
- Live global Codex reset demoed by Claire Vo at OpenAI DevDay — lennysan · 2026-09-30