今日速览
| 序号 | 标题 | 来源 | 日期 | 主题 | 推荐等级 |
|---|---|---|---|---|---|
| 1 | How to Speculate about Uncertainty in Agentic Coding? A Draft-Model Gate Method | arXiv | 2026-09-04 | AI-Agent | 高 |
| 2 | CONTINUITY: Security-Context Contracts for Composable LLM Agent Controls | arXiv | 2026-09-04 | AI-Agent | 高 |
重点论文与技术动态
1. How to Speculate about Uncertainty in Agentic Coding? A Draft-Model Gate Method
- 来源:arXiv
- 日期:2026-09-04
- 作者/机构:Konstantin Grotov, Valentin Malykh
- 主题标签:
AI-Agent,arXiv - 推荐等级:高
- 分类:cs.LG
一句话结论
该研究提出Speculative Uncertainty (SU)方法,通过小开放权重草稿模型从代理输出token中预测失败信号,显著降低软件工程代理在部署中的执行错误率和token成本。
核心内容
- SU方法无需访问代理内部参数(如logits、权重、激活或重复采样),仅从输出token恢复预测性失败信号。
- 通过分离推理和动作跨度提取相位感知特征,并校准到可验证目标生成失败概率分数。
- 实验在Qwen3-Coder-480B和Claude 3.5 Sonnet上验证,执行错误率降低6-8%,token成本减少14-19%,且泛化至分布外基准无需重训练。
方法与数据
- 方法:反转推测解码,使用小草稿模型在单次前向传播中评分代理轨迹。
- 数据:摘要未明确。
价值判断
- 值得关注:为黑盒代理提供无需内部访问的错误预测机制,避免执行后昂贵重试,提升部署可靠性。
- 可复用点:失败概率分数可被下游策略如预执行否决门、路由或人工干预直接消费。
- 局限/待核查:摘要未明确。
摘要
LLM agents deployed for software engineering fail expensively: they act confidently wrong, and bad actions are recognized only after costly execution and retry. We present Speculative Uncertainty (SU), a method that recovers a predictive failure signal for a black-box agent from its output tokens alone, with no access to logits, weights, activations, or repeated sampling. Inverting speculative decoding, a small open-weight draft model scores the agent's already-generated trajectory in a single forward pass. From these speculative cross-likelihoods we extract phase-aware features by separating the reasoning and action spans, and calibrate them against a verifiable objective. SU produces a failure-likelihood score that any downstream policy, such as routing, human intervention, or extra test-time compute, can consume directly. To show the signal is actionable, we instantiate one such policy, a pre-execution veto gate, on software engineering agents Qwen3-Coder-480B and closed-source Claude 3.5 Sonnet, cutting execution error rate by 6-8 percentage points and token cost by 14-19% in deployment, transferring to out-of-distribution benchmarks without retraining, and generalizing across agent models.2. CONTINUITY: Security-Context Contracts for Composable LLM Agent Controls
- 来源:arXiv
- 日期:2026-09-04
- 作者/机构:Chris Zheng, Geng Yang
- 主题标签:
AI-Agent,arXiv - 推荐等级:高
- 分类:cs.CR, cs.AI
一句话结论
CONTINUITY框架通过显式安全上下文契约确保LLM代理执行的安全性,防止组件边界处的安全上下文不连续问题。
核心内容
- 识别安全上下文不连续(security-context discontinuity)为安全机制组合失效的关键问题,即安全上下文可能被丢弃、扩大、反弹或重新解释。
- CONTINUITY使用假设-保证契约模型,并通过签名根授权、证明承诺等机制携带认证安全上下文,确保跨组件过渡的完整性。
- 形式化端到端后果完整性,要求每个外部效果必须由有效授权见证链接主事者、任务、来源、委托、策略状态、规范操作和最终边界。
方法与数据
- 实现参考验证器和确定性跨层故障注入套件,覆盖32种故障类别及四个应用领域。
- 在2,560个参数化攻击实例中,无有害外部效果,完成700个良性任务并提升200个模糊案例。
价值判断
- 值得关注:解决了LLM代理安全机制组合中的上下文丢失问题,提升系统整体安全性。
- 可复用点:安全上下文携带机制(如角色绑定过渡收据、效果绑定执行许可)可复用于其他安全框架。
- 局限/待核查:摘要未明确。