#AIOps
#Research
"CIBER: A Comprehensive Benchmark for Security Evaluation of Code Interpreter Agents", Feb. 2026.
// CIBER - automated benchmark that combines dynamic attack generation, isolated secure sandboxing, and state-aware evaluation to systematically assess the vulnerability of code interpreter agents against four major types of adversarial attacks: Direct/Indirect Prompt Injection, Memory Poisoning, and Prompt-based Backdoor
Post #2088
293