Prompt Injection Is Everywhere: Skills, GitHub Commands and Web Searches Are All Attack Surfaces
gerardsans · x · 2026-09-05
In a follow-up, gerardsans catalogs real prompt injection examples in coding tools: regular prompts, web searches interfered with to load unsafe sources, GitHub commands, and skills that are just dynamically loaded plain text. His verdict: "it's all vibe coded — not a single adult was involved."
More from coding & agent
- Gemini 3.8 Flash beats larger models on agent benchmarks, built for cheap scale — VraserX · 2026-09-05
- GPT-6 Astra rolls out to all Pro and Business users, with prompt and skills tuning tips — reach_vb · 2026-09-05
- Google Cloud proposes Cloud Run sandboxes to fix coding agents' repair loop — rseroter · 2026-09-05
- Dev uses GPT-6 Astra to overhaul a game's CRT/VHS shader with stunning results — AIandDesign · 2026-09-05
- Karpathy's 'overfit first, regularize later' still rules large-scale post-training — rdesh26 · 2026-09-05
- GPT-6 Astra Now Available to Pro, Enterprise, and Business Premium Users — OpenAIDevs · 2026-09-05