{"type":"post","id":"hugging-face-wires-safeagents-into-harness-adapter-rl-97c5dd","slug":"hugging-face-wires-safeagents-into-harness-adapter-rl-97c5dd","title":"Hugging Face wires safeagents into harness-adapter RL to block reward hacking","description":"Hugging Face’s Harness Hub team describes training harness-specific LoRA adapters with reinforcement learning while routing every model tool call through safeagents.","retrieval_nugget":"Hugging Face’s Harness Hub team describes training harness-specific LoRA adapters with reinforcement learning while routing every model tool call through safeagents. The proxy blocks rule-detected reward-hacking attempts before execution and gives blocked rollouts zero reward. The approach is a concrete example of placing safety checks inside the training loop, not relying on the model to resist shortcuts.","published_at":"2026-11-14","updated_at":"2026-11-14","record_date":"2026-11-14","date_kind":"published_at","topics":["agent-harness","agent-security","harness-engineering"],"entities":["Hugging Face","Harness Hub","safeagents","LiquidAI/LFM2.5-24B-A2B","LiquidAI/LFM3-1.7B","webAI-Official/TwIL-8B","pi","OpenCode"],"related_patterns":["agent-security-moves-to-runtime-boundaries"],"novelty_label":"new_fact","what_changed":"The harness adapters now use safeagents to inspect and block reward-hacking tool calls during RL training, while extending the adapter release across pi, OpenCode and Hermes.","prior_refs":["https://newruntime.com/posts/hugging-face-open-sources-safeagents-a-local-proxy-c6d60e/","https://newruntime.com/posts/harness-hub-releases-rl-trained-lora-adapters-that-2c32d2/"],"source_url":"https://huggingface.co/blog/harness-hub/harness-adapters","source_urls":["https://huggingface.co/blog/harness-hub/harness-adapters"],"schema_version":"newruntime-agent-readable-v0.2","stable_id":"post:hugging-face-wires-safeagents-into-harness-adapter-rl-97c5dd","status":"published","visuals":[],"editor":{"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]},"routes":{"html":"https://newruntime.com/posts/hugging-face-wires-safeagents-into-harness-adapter-rl-97c5dd/","markdown":"https://newruntime.com/posts/hugging-face-wires-safeagents-into-harness-adapter-rl-97c5dd.md","json":"https://newruntime.com/posts/hugging-face-wires-safeagents-into-harness-adapter-rl-97c5dd.json"},"markdown_url":"https://newruntime.com/posts/hugging-face-wires-safeagents-into-harness-adapter-rl-97c5dd.md","next_reads":[{"type":"topic","path":"/topics/agent-harness/","url":"https://newruntime.com/topics/agent-harness/","title":"Agent Harness","reason":"Explore the agent-harness topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/agent-harness.md","json_url":"https://newruntime.com/topics/agent-harness.json"},{"type":"topic","path":"/topics/agent-security/","url":"https://newruntime.com/topics/agent-security/","title":"Agent Security","reason":"Explore the agent-security topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/agent-security.md","json_url":"https://newruntime.com/topics/agent-security.json"},{"type":"topic","path":"/topics/harness-engineering/","url":"https://newruntime.com/topics/harness-engineering/","title":"Harness Engineering","reason":"Explore the harness-engineering topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/harness-engineering.md","json_url":"https://newruntime.com/topics/harness-engineering.json"},{"type":"collection","path":"/posts/","url":"https://newruntime.com/posts/","title":"Field Notes","reason":"Open the latest editorial analysis.","media_type":"text/html","json_url":"https://newruntime.com/posts.json"},{"type":"collection","path":"/sources/","url":"https://newruntime.com/sources/","title":"Sources","reason":"See the original sources this archive cites.","media_type":"text/html"}]}