Unverified50% confidenceOpinionExact time
当前前沿AI模型大多处于协作型向自主型过渡的阶段，在某些子任务上高度自主，但在完整攻击场景中仍需人类介入
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/6/2026
First Seen
Valid until: 11/4/2026
Sources
英国AI安全研究所测试结果：主流AI模型成功攻破真实系统
Related Claims
Unverified团队提出AI智能体协作存在同步范式（人在回路中）和异步范式（发射后不管）两种根本不同的范式79% similarUnverifiedMost teams currently use AI agents in a single-pass mode of 'human initiates → agent executes → human accepts.'78% similarUnverifiedMature AI application architectures typically adopt a human-machine collaborative degradation mode, in which tasks are automatically routed to human agents when AI confidence falls below a threshold, drawing on the circuit breaker pattern from microservices architecture.77% similarUnverifiedThe AI sparring partner model represents a new paradigm in which AI transitions from a task executor to a capability-enhancing collaborator.77% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/699105API
curl https://kongchang.com/api/v1/knowledge/claims/699105MCP
get_claim(id=699105)