Unverified50% confidenceFactExact time
AI agents have no discipline, only context, and everything in the context may eventually appear in logs, files, memory notes, or debug dumps.
1
Sources
50%
Confidence
Long-term
Relevance
10/11/2026
First Seen
Sources
AI智能体会泄露你的密钥：这是架构问题，而非恶意
Related Entities
Related Claims
UnverifiedAI代理并不会把整个笔记库全部读进内存，而是采用核心记忆常驻加载与技能笔记按需动态加载的两层结构79% similarUnverifiedAI models have no project memory and start each new session in an amnesiac state, so key technical decisions need to be documented in ADR files for the AI to read.78% similarUnverifiedAI coding assistants lack memory across sessions and repositories, so any accumulated context is reset to zero each time a new session is started or a repository is switched.77% similarUnverifiedTreating AI assistants as stateless tools and storing knowledge in a controllable file system rather than in volatile session context is the core principle for dealing with context loss.74% similarUnverifiedEach AI conversation is essentially a stateless reasoning process, and the AI does not truly remember the overall picture of the system.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/996756API
curl https://kongchang.com/api/v1/knowledge/claims/996756MCP
get_claim(id=996756)