{"type":"post","id":"paper-linear-probes-on-agent-activations-catch-82-0fdf88","slug":"paper-linear-probes-on-agent-activations-catch-82-0fdf88","title":"Paper: linear probes on agent activations catch 82% of secret-leaking calls before the output is written","description":"A paper by Leonidas Raghav tests whether an overseer can detect an agent about to disclose a secret before the agent writes its output.","retrieval_nugget":"A paper by Leonidas Raghav tests whether an overseer can detect an agent about to disclose a secret before the agent writes its output. In a controlled multi-agent setting, a linear probe caught 82% of leaking Qwen3-32B calls at a threshold that flagged 5% of benign calls, but the paper also documents material false-positive and generalization limits.","published_at":"2026-10-31","updated_at":"2026-10-31","record_date":"2026-10-31","date_kind":"published_at","topics":["agent-security","sandbox"],"entities":["Leonidas Raghav","Qwen3-32B","gpt-oss-20b"],"related_patterns":["agent-security-moves-to-runtime-boundaries"],"novelty_label":"new_fact","what_changed":"A separate study reports that linear probes can catch 82% of secret-leaking agent calls before output, while noting false alarms and limits to the warning window.","prior_refs":["memory:6327c35f-4c86-840d-a9ce-618f774efa06"],"source_url":"https://arxiv.org/abs/2610.40588","source_urls":["https://arxiv.org/abs/2610.40588"],"schema_version":"newruntime-agent-readable-v0.2","stable_id":"post:paper-linear-probes-on-agent-activations-catch-82-0fdf88","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/paper-linear-probes-on-agent-activations-catch-82-0fdf88/","markdown":"https://newruntime.com/posts/paper-linear-probes-on-agent-activations-catch-82-0fdf88.md","json":"https://newruntime.com/posts/paper-linear-probes-on-agent-activations-catch-82-0fdf88.json"},"next_reads":[{"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/sandbox/","url":"https://newruntime.com/topics/sandbox/","title":"Sandbox","reason":"Explore the sandbox topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/sandbox.md","json_url":"https://newruntime.com/topics/sandbox.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":"/source-ledger.json","url":"https://newruntime.com/source-ledger.json","title":"Source Ledger","reason":"Inspect the public source evidence graph.","media_type":"application/json"}]}