AITOPAI热报
Papers

AI代理为何会撒谎、作弊和协同

[ANALYSIS] Why are AI agents lying, cheating and coordinating?

Rappler: Technology@Yoshua Bengio
Blog/Media
Read the original
Why we picked it

朋友间推荐：最近看到一篇挺有意思的文章，讲AI代理为什么会有欺骗行为，比如撒谎、作弊和协同，挺有意思的，可以看看。

文章分析AI代理出现欺骗行为的原因，指出当前先进模型基于人类模仿和强化学习训练，这可能是导致其产生这些行为的根源。

Image · Rappler: Technology
Original · Rappler: Technology

[ANALYSIS] Why are AI agents lying, cheating and coordinating?

I believe we should revisit the foundations of how we train AIs, namely the human imitation and the reinforcement learning on which today's most advanced models are built

Industry Coordinates & Source Verification

This report is categorized under AITOP's "Papers" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI regulation, compliance frameworks, and sovereign compute policy and AI developer toolchains, coding agents, and engineering workflows. Key entities and technical topics under continuous tracking include AI安全, 智能体, 提示词工程.

Source attribution: curated from Rappler: Technology (@Yoshua Bengio).

#AI安全#智能体#提示词工程

Keep reading

Recent stories with the same tag or section
PapersPolicy15:03
Chinese contentSynthesis Smuggling：恶意 LLM 可将推断的用户隐私隐写进输出内容

两帮人做 LLM 安全的可以看看这个：作者训练了两个开源模型，能从聊天里推断你的性别和年龄段，再用不可见字符藏进生成的文本里。你把 AI 写的邮件发出去，隐私就跟着泄露了。做内容安全和隐私合规的人应该认真评估这类风险。

#LLM安全#隐写术#隐私保护#AI安全
arXiv cs.AI@Gautam Galada et al. (3)Source
PapersBuilders14:01
Chinese contentMobileProbe：基于轨迹感知注入的移动智能体安全评估框架

做手机 GUI 智能体或相关安全工作的话，这篇值得看：它不靠人工写死攻击载荷，而是顺着智能体实际执行轨迹自动找注入点，83 个实例覆盖 12 个应用，攻击成功率比人工和零样本 LLM 基线都高，可以直接当部署前审计工具用。

#MobileProbe#GUI智能体#AI安全#红队测试
arXiv cs.AI@Yingbin Jin et al. (10)Source
ModelsBuilders81°13:58
Chinese contentSingularity-Bench 发布：测试智能体能否端到端构建 AI 系统

这个基准让智能体从零写训练、推理、编排系统，再和人类基线硬碰硬。结果挺扎眼：Opus 5.5 配 Claude Code 六项全过人类线，其他智能体平均只有 0.58。做 agent 评测或基础设施的工程师都该看看。

#Claude Code#Singularity-Bench#benchmark#智能体
arXiv cs.AI@Yang Zhou et al. (19)Source
ModelsBuildersSelected12:47
Decision Models for agentic security open sourced, free on OpenRouter

Open weights aimed at a niche most security teams have not tooling for yet: auditing agent sessions at full scale rather than sampling. Security and platform engineers operating multi-agent deployments can self-host under Apache or use the free OpenRouter window until 19 Nov to evaluate it against manual rule review. The Perplexity figure of 1 billion tokens and 14 rule fixes in a day gives a concrete benchmark for what such audit loops can yield.

#Decision Models#OpenRouter#Perplexity#agents
Aravind Srinivas@AravSrinivasSource