Surprise Alone Learns Lies: Evidence-Gated Continual Learning for LLMs from Streaming Claims

Summary

This study examines continual learning for LLMs that select training inputs by surprise, a rule that absorbs planted falsehoods about facts the model already knows. Tested on LoRA-adapted Qwen2.5-1.5B, Qwen2.5-3B and Llama-3.2-3B students learning from streams of true and false claims, the authors find that a verifier judge removes lies only as far as it is accurate, and that the model's own prior flags contradictions with a 43% saving in judge calls. Routing contradicted claims to trusted evidence gives the lowest poison rate, but planted matching sources for 30% of lies make it worse than the judge alone.

Related items

• CriticalCVE-2026-61707: SOCFortress CoPilot focuses on providing a single pane of glass for all your security operations needs. Prior to…Similar attack · NVD/CVE Database
• LowAnthropic-Themed Phishing Emails Bypass Top Security ToolsSimilar attack · Dark Reading
• MediumCVE-2026-118760: Kubeflow Pipelines artifact routes allow cross-namespace artifact readsSimilar attack · NVD/CVE Database
• HighCVE-2026-118758: Kubeflow Pipelines path traversal in GET /artifacts/get volume handlerSimilar attack · NVD/CVE Database
• HighGHSA-w9x9-j56r-h98r: @eeacms/volto-chatbot has Unauthenticated Danswer API Confused Deputy ProxySimilar attack · GitHub Advisory Database