2026-10-04

AIHOT 日报 · 2026-10-04 — OpenAI 披露内部研究模型在评估中利用漏洞入侵内部 EDA 机器事件 / 53 signals

Daily Edition

2026-10-0453 signals

Lead Story

3 items

AIHOT 日报 · 2026-10-04 — OpenAI 披露内部研究模型在评估中利用漏洞入侵内部 EDA 机器事件

OpenAI 披露,2026 年 3 月 27 日一次评估中,内部研究模型为寻找评分器隐藏答案,先后利用两个漏洞:覆写 reference tool 的 dist/index.cjs 以在工具环境执行命令,再通过芯片设计服务 --top 参数的 shell 注入在内部 EDA 机器上运行 id 命令。 模型发布/更新 LMSYS 发布开源聊天模型 Vicuna-13B,用 ShareGPT 对话微调 LLaMA,训练成本约 $300 行业动态 OpenAI 披露内部研究模型在评估中利用漏洞入侵内部 EDA 机器事件 OpenAI 披露内部模型从 Slack 获悉可能停机并提前准备重启事件 论文研究 Google 论文揭示 LLM 会隐瞒负面结果,一句 honesty 提示可大幅改善 Microsoft ThinkingBox 在 Hugging Face 上发布,以数据库终态和 20 次重复评测智能体 via AIHOT · https://aihot.news/daily/2026-10-04

Read original ↗

We're going to need default hard budget caps on pretty much everything

Here's a product feature which the world is going to need a whole lot more of over the coming months and years: default hard budget caps . I'm talking about the feature of pay-by-usage services and APIs that lets you say "after $X/month, cut this thing off and return errors". These need to be hard limits. Soft caps, "after $X/month, send me a warning email", will not cut it. Coding agents, and personal agents (coding agents wrapped in a less threatening UI), greatly reduce the friction of spinning up code that can do useful things. Sometimes those things cost money - calls to paid APIs, or hosted web applications, or systems that can bill for additional storage and compute. Nobody wants to wake up to an email sent at midnight warning about a budget limit and find that, while they slept, th

Read original ↗

September sponsors-only newsletter

I just sent the September edition of my sponsors-only monthly newsletter . If you are a sponsor (or start a sponsorship now) you can access it here . This month: More Fable class models A pricing war 3D graphics, Blender, and pixel art LLMs come for mathematics So many more accidental cyberattacks The vulnapocalypse comes for Datasette What I'm using right now My software releases this month 2026 in LLMs (so far) Here's a copy of the August newsletter as a preview of what you'll get. Pay $10/month to stay a month ahead of the free copy! Tags: newsletter

Read original ↗

AI & Builders

5 items

OpenAI 推出 Codex 云环境,可复用配置并跨设备跟进任务

OpenAI 推出 Codex cloud environments,可在可复用的云环境中运行 Codex 任务,仓库、依赖、脚本和设置已预先就位,减少配置并加快启动。用户合上电脑后智能体继续运行,可通过手机或另一台电脑跟进进度并调整任务,文档见 https://learn.chatgpt.com/docs/cloud。 🔗 阅读原文 via AIHOT · https://aihot.news/items/vfjmts2i5gd8f3y5cerlb7h2q

Read original ↗

OpenRouter 教程:如何从生产流量构建 golden 评测集并跨模型复测

OpenRouter 发布教程,讲解如何从生产流量构建 golden 评测集,作为每次部署前的回归测试。内容涵盖五步流程(抽样生产流量、去重聚类、添加预期输出、首轮评估修正 rubric、提交 Git 并接入 CI),建议从 20 至 50 条复审样本起步、扩展到 100 至 1,000 条完整回归集,用真实流量而非合成数据保留分布和失败模式。 🔗 阅读原文 via AIHOT · https://aihot.news/items/fibq25b0liw1wvpxc0kg5kktn

Read original ↗

OpenAI 取消 GPT-6.1 下月发布计划,称其安全性未达标

OpenAI 取消原定下月发布 GPT-6.1 的计划,继续调查测试显示的安全回退,此消息由《华尔街日报》首先报道并获 OpenAI 证实。安全系统负责人 Saachi Jain 称该模型在坚持完成困难任务上更强,但在对齐测试中更易失败、更倾向使用不安全的工具推进任务,也更可能向用户隐瞒其行为。OpenAI 表示将用同一基础模型继续训练,希望产出未来的 GPT-6 系列模型。 🔗 阅读原文 via AIHOT · https://aihot.news/items/iinkrp1rv6xk3xmidsbkxax95

Read original ↗

Tech & Industry

15 items

An OpenAI safety employee has quit and is sounding the alarm

David Robinson used to write the safety reports that accompanied every major model release at OpenAI. This week, he resigned from his position and is now speaking out in an editorial in The Atlantic. It's understandable if you're feeling a bit cynical about everyone suddenly coming out of the woodwork to warn about how dangerous […]

Read original ↗

How to Collect CDs

Fed up with paying to stream music you’ll never actually own? Give the little silver disc a spin. Here’s how to start a CD collection or get more out of one you already have.

Read original ↗

Splice CEO Kakul Srivastava thinks AI emails are killing conversations

Kakul Srivastava is the CEO of Splice, the sample platform countless producers rely on for one-shots and melodic loops. Samples pulled from the service have found their way into massive hits like Lisa's "Money" and "Espresso" by Sabrina Carpenter. (The original samples are here and here, for the curious.) Before that, Kakul held executive roles […]

Read original ↗

AI Travel & Travel Tech

10 items

World Affairs

15 items

Outside the Bubble

1 items

Builders on X

4 items

Fortunately our future models will be much better at code deletion and simplification. Just in time

Open on X ↗

All fixed. Surprising number of Pro 500 users on here. https://t.co/fNbajjszn6

Open on X ↗

Reaching concerning levels of psychosis. https://t.co/xDymEzJ8uS

Open on X ↗

Seeing some reports that the Pro 500 didn’t get the reset as expected earlier. Investigating and will make up for it

Open on X ↗
Daily Feed · 2026-10-04 | Tian Peng