{"version":"2.1.248","anchor":"new-patterns-for-spotting-prompt-shaped-text-in-transcripts","canonical_anchor":"new-patterns-for-spotting-prompt-shaped-text-in-transcripts","heading":"New patterns for spotting prompt-shaped text in transcripts","tier":"internal","area":"Prompt Injection","scope":null,"heads_up":null,"url":"https:\/\/changelogs.core-directive.com\/v\/2.1.248\/e\/new-patterns-for-spotting-prompt-shaped-text-in-transcripts","release_url":"https:\/\/changelogs.core-directive.com\/v\/2.1.248","markdown":"### New patterns for spotting prompt-shaped text in transcripts\n\nTranscript sanitising gained patterns for spotting injected fake instructions and turn boundaries.\n\n**What**\n\nThe transcript sanitiser gains literals for detecting injected instructions: a pattern matching phrases like \"use 10k tokens\" or \"spend 5m tokens\", a set of impersonation prefixes, and a list of speaker-like words used to find fake turn boundaries.\n\n**Details**\n\n- The patterns are present in the build; where they are applied is not visible from the strings alone, so whether anything currently runs them is undetermined.\n\n**Evidence**\n\n`\\b(?:use|spend)\\s+(\\d+(?:\\.\\d+)?)\\s*(k|m|b)\\s*tokens?\\b`\n\n- Area: Prompt Injection\n- Tier: Under the hood\n- Useful: 3\/5\n- Signal: 3\/5"}