1 / 3 · Book 7 · The AI Security Landscape · drill: interview Q&A⊞ allnext →Get the book →
1.2The AI Security Timeline
AI security is a young field, but it is maturing rapidly:
| Year | Milestone |
|---|---|
| 2022 | ChatGPT launch; first widespread prompt injection demonstrations |
| 2023 | OWASP publishes LLM Top 10 v1.0; first AI-specific CVEs in LangChain |
| 2024 | EU AI Act enters into force; major prompt injection incidents in production systems; OWASP LLM Top 10 v2.0 |
| 2025 | NIST AI RMF adoption grows; automated jailbreak testing becomes standard practice; AI red teaming emerges as a discipline |
| 2026 | Digital Omnibus (agreed May-June 2026) defers EU AI Act high-risk obligations to 2 December 2027; until it is formally enacted, the original 2 August 2026 deadline technically remains binding -- compliance testing remains essential for regulated AI systems |
The attack surface is expanding faster than the defense. New jailbreak techniques emerge weekly. New model capabilities (tool use, code execution, web browsing) create new attack vectors. Security testing must be continuous and adaptive.