Your AI agent's secrets leak because you fed API keys straight into the LLM
sujingshen · x · 2026-10-14
Venice AI's Joshua Mo says most AI agents hand passwords and API keys directly into the LLM's context, where they leak via training data, prompt injection, malicious tool calls, or observability pipelines that log prompts by default. His fixes: keep secrets in a deterministic execution layer, store hashed identifiers instead of raw data, use split-key custody for encrypted payloads, and demand hardware attestation of private inference. He concedes internet-enabled agents can still exfiltrate data via external APIs. The actionable takeaway: audit credential flow and logging defaults, not the model name.
More from coding & agent
- Schmidhuber team unveils JevGPT dual-system agent for faster, cheaper autonomous research — _reachsumit · 2026-10-14
- Airbnb case study: portable agent skill hits 99.2% procedural accuracy in 1-2 engineer days — _reachsumit · 2026-10-14
- Ship faster sites: have Opus send 4 subagents to audit 20 performance bottlenecks before launch — MartinGTobias · 2026-10-14
- Claude Code makes a SpaceX documentary video entirely via chat, no editing software — xiaohu · 2026-10-14
- Copilot Cowork review: a year behind rivals, riddled with UX gaps — koltregaskes · 2026-10-14
- Agent told to change one class time rewrote the waiver and reverted prices — StormController · 2026-10-14