Unverified50% confidenceTradeoffExact time
模型能力越强、越能调用外部工具，其潜在攻击面就越广，能力扩张与安全防护存在内在张力
1
Sources
50%
Confidence
Long-term
Relevance
7/20/2026
First Seen
Sources
Related Claims
UnverifiedSkills机制中工具调用引入大量外部数据，使间接注入攻击风险尤为突出80% similarUnverifiedLarge model-related skills are gradually evolving from a bonus qualification to a hard requirement.70% similarUnverifiedDefense in depth is currently the most pragmatic strategy for addressing prompt injection, involving the layering of security measures across multiple levels including input, model, output, and application.70% similarUnverifiedAs the capabilities of underlying models converge, the ease of use and efficiency of developer tools will become key competitive dimensions.65% similarUnverified无载荷技能攻击通过技能的元数据、描述文本、命名结构或调用逻辑的组合，诱导智能体在特定上下文中触发非预期的危险行为65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/566651API
curl https://kongchang.com/api/v1/knowledge/claims/566651MCP
get_claim(id=566651)