Summary
The IJCAI-ECAI challenge focuses on prompt injection, skill poisoning, supply-chain vulnerabilities and other threats affecting agents with system-level operations. It brings offensive and defensive techniques together in a common evaluation setting.
Why it matters
Agent security cannot be assessed only through model refusal tests because deployed agents inherit risks from tools, permissions and third-party skills. Shared attack-and-defence benchmarks can make those risks more measurable.
Related coverage
- RobustifAI targets deployment failures beyond benchmark scores
- Travel-planning agents face a constraint-heavy evaluation
- TRUST-AI convenes on trustworthy AI