AI Risk Crossword

Edit Answers
12345678910111213141516171819
Across
  1. 1. To attack your own AI to find its holes (one word)
  2. 3. Abused credentials let an agent act beyond its scope (ASI03)
  3. 4. A confident fact an LLM just made up
  4. 7. Insurance for AI agents that lapses on compliance drift
  5. 8. The buried clause that says "not covered" — the AI gap incumbents dodge
  6. 9. Prompt trick that bypasses a model's safety rules
  7. 12. When the hidden system prompt spills to the user (LLM07)
  8. 13. A safety filter wrapped around a model's input and output
  9. 15. NIST AI RMF function that sets the risk culture
  10. 17. Tricking an agent into abusing its own connected ones (OWASP ASI02)
  11. 18. Planting false data in an agent's store to warp later decisions (ASI06)
  12. 19. US agency behind the AI Risk Management Framework
Down
  1. 2. Confident, wrong output a model propagates as fact (LLM09)
  2. 5. Corrupting training data or weights to plant bad behavior (LLM04)
  3. 6. Too much autonomy or permission handed to an LLM (LLM06)
  4. 10. MITRE's adversarial-ML attack knowledge base
  5. 11. A misaligned agent acting on its own hidden agenda (ASI10)
  6. 12. The legal hook for who pays when the AI causes harm
  7. 14. Hidden text in a prompt that hijacks the model's behavior (OWASP LLM01)
  8. 16. Body behind the Top 10 lists for LLM and agentic apps