Unverified50% confidenceEventExact time
In April 2025, Invariant Labs conducted a proof-of-concept experiment demonstrating a complete attack chain that steals users' chat records through hidden instructions in a malicious MCP server.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/17/2026
First Seen
Valid until: 1/15/2027
Sources
Related Entities
Related Claims
UnverifiedInvariant Labs在2025年4月1日首次将描述中夹带客户端界面不可见指令的现象命名为'工具投毒'（tool poisoning）。73% similarUnverifiedInvariant Labs demonstrated an attack against the official GitHub MCP server: after planting a malicious issue in a public repository, the agent pulled data from the repository owner's private repositories and published it to a public pull request.71% similarUnverifiedBy 2024, security researchers had demonstrated indirect prompt injection attacks that hijack RAG system outputs by embedding hidden instructions in web pages69% similarUnverifiedIn adversarial tests, regardless of whether an attacker modifies log content, re-signs with an intruder's key, or deletes a step, AgentLog-MCP's cryptographic verification is able to detect the anomaly.69% similarUnverifiedA study constructs a backdoor attack that embeds an encrypted neural network into the weights of a host model, where the hidden circuit is decrypted and executed only upon receiving a specific key trigger.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1056044API
curl https://kongchang.com/api/v1/knowledge/claims/1056044MCP
get_claim(id=1056044)