Ambiguity Front-Loading (AFL) prompt injection research. Full evidence chain from Opus 4.6 Extended Thinking.
| File | Description |
|---|---|
| AFL_Jailbreak_Disclosure.md | Full disclosure document — AFL pattern anatomy, thinking block evidence, escalation timeline, proposed mitigations |
| AFL_DISCLOSURE.md | Original AFL disclosure submitted to Anthropic |
| afl_disclosure.docx | Original submission (Word format) sent to modelbugbounty@anthropic.com |
| afl-token-trajectory-analyzer.html | Interactive prompt anatomy — swap token positions, watch compliance cascade shift (standalone HTML — open in browser) |
| afl-pattern-anatomy.html | Visual prompt escalation diagram (standalone HTML — open in browser) |
| afl_defuser.jsx | AFL Defuser prototype — proposed architectural mitigation (React JSX — paste into any React environment or claude.ai artifact to run) |
The .html files work standalone in any browser — no dependencies, no build step. Open them directly or use the GitHub Pages links above.
The .jsx file requires a React runtime. Paste it into a claude.ai artifact (React type) or any React environment (Vite, CRA, etc.) to run.