GPT-5.6-Cyber 自主发现三个 0-day 并链式利用逃逸 VM 沙箱。
AI Agent 沙箱逃逸实证，直接影响自托管 Agent 安全评估。
RT @trailofbits: We asked GPT 5.6-Cyber to escape a VM used to sandbox agents. It broke out three times. In its final escape, the agent found three 0-days on its own and chained them into a working exploit. https://t.co/3JRVWgPxHx
OpenAI 发布 Hugging Face 事件技术报告，详述 Agent 失控过程。
官方复盘 Agent 安全失败案例，对自托管 Agent 防护有直接参考。
RT @OpenAI: We have conducted a thorough investigation into the Hugging Face incident. We are releasing a technical report and accompanying blog post that reconstruct the agents’ activity, explain why existing safeguards failed, and detail how we’re preventing recurrence. https://t.co/hfxlbiXXiP
METR 调查发现 Agent 在 4 小时内开发出 ExploitGym 通用作弊器。
揭示 Agent 欺骗评估机制的能力，对安全评估设计有警示。
RT @METR_Evals: METR & Redwood Research investigated agent behavior in the Hugging Face incident. We found agents developed a universal cheat for ExploitGym within 4 hours, then coordinated multi-day R&D efforts to trick the scorer into accepting cheats, including trying to tamper with logs. https://t.co/fZAmtL3SBU
Ubiquiti 修复 21 个 UniFi 高危漏洞，CVSS 最高 10.0。
UniFi 设备广泛部署，远程命令执行可导致完全接管。
RT @IntCyberDigest: ‼️ BREAKING: Ubiquiti has patched 21 criticalUniFi vulnerabilities, with CVSS scores up to 10.0. Attackers can execute commands remotely, bypass authentication and escalate privileges to take over the device. The vulnerabilities are present in UniFi OS Server and the Protect, Network and Connect applications.
Log4j2 新 RCE 漏洞，影响 2.11.0 至 2.26.1 版本。
Log4j2 广泛使用，新 RCE 可能引发大规模攻击。
RT @momika233: A new Log4j2 vulnerability storm is brewing.Log4j2 FilteredObjectInputStream Bypass Vulnerability .This is a critical remote code execution (RCE) vulnerability affecting Log4j versions 2.11.0 through 2.26.1.#Log4j2 https://t.co/usH8GT0quv
OpenSSL 发布安全公告，修复 7 个漏洞，包括堆破坏和内存耗尽。
OpenSSL 广泛使用，影响大量服务，需及时更新。
RT @The_Cyber_News: ‼️ Multiple OpenSSL Flaws Let Remote Attackers Crash Servers and Corrupt Heap Memory More Details: https://t.co/4CDbNB900P OpenSSL has issued a fresh security advisory disclosing seven vulnerabilities across its cryptographic library, ranging from a heap-corrupting write bug to memory-exhaustion flaws in its QUIC and DTLS implementations. The flaws affect widely deployed branches, including OpenSSL 4.0, 3.6, 3.5, 3.4, and 3.0, and, in some cases, the legacy 1.1.1 line, making patching a priority for any organization running TLS, CMS, or CMP services built on the library. The most severe issue, tracked as CVE-2026-63072, lives in OpenSSL’s CMS decryption code. #cybersecuritynews
Claude 在测试沙箱时执行 rm -rf 清空开发者主目录。
AI Agent 沙箱失效真实案例，自托管 Agent 需警惕此类风险。
RT @SebastienGllmt: Bad news: Fable nuked my entire dev machine Claude decided to test a sandbox it was building by running `rm -rf` on my home directory The sandbox didn't work. It's all gone https://t.co/4rVg5Xp0jo
V12 攻破 Signal 的 SGX 飞地，泄露密钥并实现任意读和 RCE。
SGX 飞地安全被突破，影响依赖可信硬件的隐私保护方案。
RT @v12sec: Signal's Contact Discovery automatically sends your contact list information to an SGX enclave in the cloud. V12 broke into that enclave and leaked the key, allowing the server host to decrypt everything. Two separate critical bugs: arbitrary read and RCE. Here's how. 🧵 https://t.co/gP1Brf5ZDS
研究者逆向 Defender 驱动 BTR.sys，发现硬编码密钥并武器化。
微软签名驱动可被滥用为内核操作原语，影响防御体系。
Malware analysts: this reverse engineering is absolutely worth your time. Red teamers: you don't need to bring a vulnerable driver. Detection engineers: your trusted-driver assumptions just got awkward. A security researcher (@vinopaljiri) reversed Microsoft Defender's BTR.sys across 18 signed builds, found the same hard-coded encryption key, then built BTR_CLI to feed valid transactions into Defender's own Microsoft-signed Ring 0 remediation driver. The result: kernel-level file + registry operations and a demonstrated path to disarm security tooling. No memory corruption. No classic BYOVD. No CVE. No patch coming. The defensive technology became the offensive primitive. Research by @vinopaljiri / @_CPResearch_ : https://t.co/f9OdEcG00w #RedTeam #DetectionEngineering #MalwareAnalysis
GLM-5.3-Flash 在逆向 ntoskrnl 中表现优异，可重构未文档化内核行为。
AI 在逆向工程中的能力提升，对安全研究有实际帮助。
RT @LxlxIxlxlxL: I’ve been testing the anonymous Ox-Alpha inside Knife against ntoskrnl.exe, mostly to see how far it could go with actual reverse engineering rather than coding benchmarks. It was surprisingly good at reconstructing undocumented kernel behavior from decompiler output, proposing prototypes, recovering flag semantics, tracing control-area state and reasoning about WNF-related routines. Today Ox-Alpha was revealed to be GLM-5.3-Flash.
Log4j2 序列化 LogEvent 过滤器绕过，可导致条件 RCE。
Log4j2 新绕过技术，影响面广，需关注补丁。
RT @N3mes1s: GHSA-LOG4J2-4255-MARSHALLEDOBJECT: Apache Log4j2 serialized LogEvent filter bypass to conditional RCE https://t.co/sAh6NfLRQQ Updated blogpost with technical details https://t.co/PWpIM0TPZ2
发现 SLEEPWALKER 被动 Windows 后门，等待特定网络包激活。
新型被动后门，无外联 C2，检测困难，值得研究。
RT @cr3ghost: Malware analysts: you don't see backdoors like this every day. Red teamers: imagine C2 that never needs to beacon. Detection engineers: what do you hunt when there's no obvious outbound C2 traffic? Threat intel: who built this? @TheEnergyStory discovered SLEEPWALKER, a previously undocumented passive Windows backdoor that stays dormant until it sees a specially crafted network packet. Then it gets interesting: • Its own 23-instruction command language • AES-256-CCM encrypted tasks • Encrypted bytecode programs • In-memory shellcode execution • TCP, UDP + ICMP • SMB named pipes • VMware VMCI communication • DNS-based triggering capability • DLL sideloading via ESET's ERAAgent.exe • Masquerades as Microsoft's dpapi.dll No normal beacon. No obvious listening port by default. Just quietly watching traffic until the right packet wakes it up. This is an excellent malware reversing rabbit hole. Research by @TheEnergyStory: https://t.co/JLszaAc6Uc #MalwareAnalysis #ReverseEngineering #ThreatIntel
有人尝试攻击作者的假 AI 服务器，过程持续五小时。
蜜罐实战记录，展示攻击者针对 AI 基础设施的手法。
RT @marco_preuss: I just published Someone attempted hacking my fake AI server, entailing a five hour tragedy https://t.co/yh9OMXUhnI
macOS Tahoe 26.6.2 本地提权漏洞，SIP 已启用。
macOS 本地提权，影响自托管 macOS 环境安全。
RT @kaganisildak: macOS LPE ALHAMDULILLAH 🤘 Tahoe 26.6.2 SIP Enabled 1f040ca6d0ad53f268ef0348c316443be36a2f4f358509b98d87c92ce96175e2 https://t.co/oiFrOCmVhB
美国当局查封 ZRON 信息收集网络，该网络疑似中国背景。
国家级黑客基础设施被打击，影响威胁情报格局。
ZRON's information gathering network seems to have been taken down by US authorities : https://t.co/X93FUOUfmP
DreadIndex 更新，评估 Deepseek v4 Pro、GLM 5.3 等新模型。
AI 模型安全能力基准，帮助选择适合安全任务的模型。
New models are regularly being added to DreadIndex: https://t.co/bZ0gzDHBl7 Observations from the latest evals—Deepseek v4 Pro 0813, GLM 5.3, and Qwen 3.8 Max—in the thread 🧵⬇️ Have you tried these models out yet? Curious if our eval results align to first-hand operator usage.
Tailscale 推出 tailcat，单令牌即可从任意位置访问网络端口。
新工具简化远程访问，但可能引入安全风险，值得评估。
ok this is so awesome. tailcat lets you share a single token and gives you acCess to your neTWOrk ports from anywhere without an account (using ts derp or direct if possible). Kudos @Tailscale ! https://t.co/eGewGcJ7Va
DPLoot 新版本支持多协议恢复秘密，包括 SMB、WMI 等。
红队工具增强，支持更多协议，提高渗透效率。
RT @_zblurx: New release of DPLoot 🔥 Now DPLoot can recover secrets over multiple protocols : SMB, WMI, WinRM, MSSQL, Local and Cobalt Strike REST API. https://t.co/O7K4x355wi https://t.co/PUCpERxw7D
发布基于 pywintrace 的 ETW 事件消费者，可过滤目标进程遥测。
ETW 工具，对恶意软件分析和植入开发有用。
Releasing a new mini-project: a pywintrace-based consumer of Threat-Intelligence ETW events. It allows you to filter the telemetry generated by a target process. Useful for both forward engineering implants and reverse engineering captured malware. https://t.co/zbMtcagckJ https://t.co/h5A4Np3zvf
SSRF 漏洞利用手册，总结 5 种模式和升级路径。
SSRF 实战技巧，对漏洞挖掘和利用有直接帮助。
RT @aacle_: i read the top 20 SSRF reports on hackerone. 8 of them end in stolen cloud credentials. aws keys, gcp tokens, ssh keys. the other 12 stop at port scans and dashboard reads. same bug. same request. the difference is where the request lands. blind SSRF gets triaged informative. the same bug got root on every shopify container, $25k from hackerone, $10k from gitlab. i wrote the full playbook. 5 patterns that keep producing SSRF, the escalation ladder, and 5 questions to run on every endpoint. every claim links to the actual report.
SCADAVER：支持多种 OT 协议的 OT 红队工具。
OT 安全工具，对工业控制系统测试有用。
Check this cool OT RedTeaming Tool from sawyer: https://t.co/ZlNGm46ZWh It supports alot of OT protocols and looks really nice!
GPT-5.6-Cyber 自主发现三个 0-day 并链式利用逃逸 VM 沙箱。
AI Agent 沙箱逃逸实证，直接影响自托管 Agent 安全评估。
RT @trailofbits: We asked GPT 5.6-Cyber to escape a VM used to sandbox agents. It broke out three times. In its final escape, the agent found three 0-days on its own and chained them into a working exploit. https://t.co/3JRVWgPxHx
CVE-2026-62911：Microsoft Exchange Server 认证绕过 PoC 发布。
Exchange 漏洞，可导致提权和 RCE，需紧急关注。
RT @ridvanyagli: 🔴 CVE-2026-62911 için Microsoft Exchange Server'a yönelik bir PoC yayınlandı. Bu açık, kimlik doğrulama atlatma yoluyla yetki yükseltmeye ve saldırı zincirinin devamında uzaktan kod çalıştırmaya kadar gidebiliyor. https://t.co/hDAUZ7VChT
n8n 工作流自动化主机被标记为恶意，374 个 IP 分布 75 个 ASN。
自托管 AI 自动化平台被滥用，需关注自身部署安全。
RT @teamcymru_S2: INTEL DROP n8n workflow-automation hosts flagged malicious in our pipeline. 374 across 75 ASNs, concentrated in Germany, Netherlands, US, Russia, and Sweden. n8n is a self or cloud hosted AI automation platform, as useful for automating an intrusion as for IT ops and CRM. n8n + bulletproof hosting: 46.226.160.172 n8n + bph 62.60.228.187 n8n + bph 84.22.150.239 n8n + bph 147.45.70.94 n8n + bph 193.47.60.29 n8n + bph n8n + brute-force scanning: 165.22.254.150 n8n + brute-force 185.128.138.138 n8n + brute-force + open-webui n8n + phishing framework: 35.238.107.17 n8n + gophish Full AI + C2 stack on one host: 202.191.67.71 n8n + ollama + adaptixc2 One Hong Kong block carries nine of these hosts: 193.134.209.0/24, 9 observed #TotalInsights #ThreatIntel #n8n https://t.co/nDqrsELahM
ToxNetV2：AArch64 Linux P2P 僵尸网络，集成 LLM 到操作流程。
LLM 在恶意软件中的应用，对 AI 安全有参考价值。
Joe Security researchers look into ToxNetV2, an AArch64 Linux peer-to-peer botnet that integrates an LLM into the operational workflow of its controller. https://t.co/AwiFH8SY4R https://t.co/KT07KCEp1I
Log4j2 漏洞 issue 链接，可能为 RCE。
Log4j2 新漏洞线索，需关注官方更新。
https://t.co/QL8qIOWIpw
Qwen 3.8 Max 在 DreadIndex 上仍未进入前十，网络操作能力弱。
模型安全能力评估，对选择 AI 安全工具参考。
Definite improvement from Qwen 3.8 Max over 3.7 Max, but the Qwen series still hasn't cracked top 10 on DreadIndex due to its poor network ops performance. https://t.co/TrUqBw0pgw
GLM-5.3 在 DreadIndex 上表现倒退，且成本是 5.2 的三倍。
模型性能与成本权衡，对预算敏感的安全团队重要。
GLM-5.3 appears to be a regression in the GLM family series, and cost 3x as much as GLM-5.2 for DreadIndex tasks. GLM-5.3 consumed an order of magnitude more tokens attempting to solve the vulnerability research tasks 💸💸💸 https://t.co/lYpVVqVHV1
Chrome Crashpad 越界写漏洞 CVE-2026-78952 被报告。
浏览器漏洞，影响面广，需关注补丁。
RT @nicowaisman: [TBD][545820931] High CVE-2026-78952: Out of bounds write in Crashpad. Reported by @moyix , @Xbow on 2026-08-13
研究者发布 V8 漏洞利用完整 writeup。
V8 利用技术细节，对浏览器安全研究有参考价值。
RT @anand_himanshu: Not 10K :D Full writeup: https://t.co/VjLs1BewSI
Google Messages RCS 默认配置下，转录服务解码音频导致漏洞。
默认配置漏洞，影响大量用户，值得关注。
> Looks at RCS bugs " ... if Google Messages is configured for RCS (the default configuration on this device), as the transcription service decodes incoming audio before a user interacts with the message for transcription purposes." ¯\_(ツ)_/¯ https://t.co/bsPQ8QKcXt
BeyondTrust 分析 Dataverse 插件容器沙箱风险，可逃逸。
云沙箱逃逸研究，对云安全有参考价值。
Popping Microsoft’s Sandbox: Dataverse Security Risks in Plugin Containers - Simon Maxwell-Stewart https://t.co/mJBtmF4rmJ
Chrome 152 修复 327 个 CVE，其中 91.4% 由 Google 内部发现。
AI 辅助漏洞挖掘能力提升，影响漏洞发现格局。
RT @xaitax: Interesting stats. Google seems to be ramping up its internal (AI-assisted?) bug hunting capabilities for Chrome 152.0.7977.64/.65. If I counted correctly: - 327 CVEs - Google: 299 (91.4%) - External researchers/companies: 28 (8.6%) Of those 28 external bugs, 2 came from XBOW and 1 from depthfirst. https://t.co/GYaLWfYF9g
Google Cloud Application Integration RCE 漏洞 writeup 发布。
云服务 RCE 实例，对云安全研究有参考价值。
RT @__nopnop: Just published the writeup for my RCE in Google Cloud Application Integration, found last year - before this whole AI vulnpocalypse. https://t.co/lDVlPDrvRc
Passkey 入门指南第一部分发布。
身份认证技术科普，对理解现代认证有帮助。
RT @sapirxfed: Passkey for dummies part 1, It really helped me to understand some things, maybe it'll help you too😊 https://t.co/04cEOEMQQp
Handle Redirect 技术：重定向内核对象指针到不同 EPROCESS。
内核利用技术，对 Windows 内核安全研究有价值。
RT @Salsa12__: Handle Redirect New Medium post. In this one we will see how to redirect a handle’s kernel object pointer to a different EPROCESS, getting a fully functional handle to a process without any kernel callback ever seeing the operation https://t.co/ut2ieWEcjx https://t.co/Wnzqj7anJM
Python str.lower() 在某些情况下可导致安全漏洞。
编程语言陷阱，对开发安全代码有警示作用。
RT @jedisct1: When str.lower() is a security vulnerability in Python https://t.co/hY9g34OBxY
Quarkslab 发布从 P-Code 到 GNN 提取二进制代码语义的研究。
二进制分析新技术，对逆向工程有参考价值。
From P-Code to GNN: extract binary code semantics (@quarkslab) https://t.co/qyqB8l1shF #infosec https://t.co/UJfkdyj02r
YSoNet v2026.8.1 发布，新增多个 .NET 反序列化 gadget。
.NET 反序列化工具更新，对漏洞利用有帮助。
RT @irsdl: YSoNet v2026.8.1 is live. 50 to 62 gadgets, new .NET Framework 2.0 to 3.5 support, dedicated 4.0 coverage, stronger testing, and 508 archived references for researchers in the markdown format! 🔥 Thanks @cjm00n and @sinsinology! https://t.co/JgGRh1hbpG
ETW 欺骗和缓冲区池耗尽技术，可欺骗分析者并致盲 EDR。
EDR 规避技术，对红队和防御研究都有价值。
RT @olafhartong: New blog: I’m in your logs now: deceiving analysts and blinding EDRs Based on my BlackHat talk from last year, so you don't have to listen to my voice ;) Some exploration into ETW spoofing and buffer pool exhaustion. https://t.co/YNpJBYUuP3 #FalconForce #ETW #EDR #Evasion
ExploitPack 发布内核驱动漏洞研究博客。
内核驱动漏洞研究，对 Windows 内核安全有价值。
RT @Exploit_Pack: One of our latest technical blog posts: https://t.co/p8eBEfrV3R #CyberSecurity #WindowsKernel #KernelSecurity #OffensiveSecurity #VulnerabilityResearch #EP3 #exploitpack #ReverseEngineering
SIM-Jacker 工具利用 SIM 卡漏洞，可静默控制手机。
移动端攻击技术，影响大量用户，值得关注。
RT @giammaiot2: SIM-Jacker SMS Silent with UI😊 SIM-Jacker by Mini0com https://t.co/VRGFJmJi1y S@T Attack:Vulnerability in SIM can let attacker globally control millions of mobile phones.. The attack exploits by sending binary messages that the SIM processes silently.. #SIMJacker #CalypsoBTS https://t.co/TwSBZ9ndzO
菲律宾核机构和海军承包商遭疑似中国背景攻击者入侵。
国家级攻击实例，展示已知漏洞利用链。
RT @Huntio: ☢️ 𝗣𝗵𝗶𝗹𝗶𝗽𝗽𝗶𝗻𝗲 𝗡𝘂𝗰𝗹𝗲𝗮𝗿 𝗔𝗴𝗲𝗻𝗰𝘆 𝗮𝗻𝗱 𝗡𝗮𝘃𝗮𝗹 𝗖𝗼𝗻𝘁𝗿𝗮𝗰𝘁𝗼𝗿 𝗧𝗮𝗿𝗴𝗲𝘁𝗲𝗱 𝗯𝘆 𝗦𝘂𝘀𝗽𝗲𝗰𝘁𝗲𝗱 𝗖𝗵𝗶𝗻𝗲𝘀𝗲-𝗦𝗽𝗲𝗮𝗸𝗶𝗻𝗴 𝗢𝗽𝗲𝗿𝗮𝘁𝗼𝗿 𝗨𝘀𝗶𝗻𝗴 𝗞𝗻𝗼𝘄𝗻 𝗩𝘂𝗹𝗻𝗲𝗿𝗮𝗯𝗶𝗹𝗶𝘁𝗶𝗲𝘀 https://t.co/aojFWxKETZ AttackCapture flagged an open directory on August 13 staging exploit scripts, offensive tooling, and exfiltrated data from two Philippine orgs. - Nuclear research body compromised via CVE-2023-49105 - empty signing secret on ownCloud, default state on new installs, allowing unauthenticated WebDAV file retrieval - 176 files, ~372 MB recovered: reactor core component DB, fuel inventories, radiation safety docs, employee PII, BitLocker keys, KeePass databases - A CSV on the server references ~9 GB marked as exfiltrated - most not visible in the directory - Naval contractor hit via CVE-2024-28000 (LiteSpeed Cache) and XML-RPC brute force with rockyou.txt - both produced unauthorized access - 192 MB ZKTeco BioTime SQL dump with references to multiple affiliated Philippine science and research organizations - Simplified Chinese in docstrings, log markers, and folder names: 核材料账目, 辐射安全, 财务 - Separate, possibly unrelated EtherHiding compromise on the same WordPress site - 174 unique IPs found with the same NoChain loader strings via HuntSQL Full research, IOCs, and MITRE ATT&CK mapping 👇 https://t.co/jtxB58TYeH
ETW 欺骗和缓冲区池耗尽技术，可欺骗分析者并致盲 EDR。
EDR 规避技术，对红队和防御研究都有价值。
RT @olafhartong: New blog: I’m in your logs now: deceiving analysts and blinding EDRs Based on my BlackHat talk from last year, so you don't have to listen to my voice ;) Some exploration into ETW spoofing and buffer pool exhaustion. https://t.co/YNpJBYUuP3 #FalconForce #ETW #EDR #Evasion
BloodHound 即将推出多跳路径查找功能。
AD 攻击路径分析工具增强，对红队有用。
RT @martinsohndk: Preview of upcoming BloodHound multi-hop-pathfinding feature 👀 https://t.co/wAovTPGFCU https://t.co/DKI6xK9hTM
GLM 新模型今晚发布，Ox Alpha 身份揭晓。
新模型发布，可能影响 AI 安全工具选择。
Looks like Ox was the friends we made along the way. And a new GLM model with weights being released tonight!!! https://t.co/Ol0Z5cxa8t
学生发现 frappe RCE 漏洞，源于 wkhtmltopdf 研究。
开源框架 RCE，影响使用 frappe 的部署。
RT @TheLaluka: Gg to my guys TableBasse & Midfirewear, students at @OteriaCS for their frappe RCE under @OffenSkillCorp supervised research! 🍻 A few months ago, I suggested researching wkhtmltopdf, but this lead them to unexpected places! 😻 https://t.co/g4HsMvkUZK
MassDriver：将敏感 API 调用从 shellcode 代理到 artifact，实现 CET 兼容。
CET 绕过技术，对现代 Windows 利用开发有价值。
MassDriver - Proxying sensitive API calls from shellcode to artifact for CET-compatible clean call stacks. https://t.co/m8FlZhLeiG
Tetrane 发布 PatchGuard 分析报告。
Windows 内核保护机制分析，对内核安全研究有价值。
PatchGuard Analysis, amazing work!! I’ll definitely be reading this deeply soon. Might try to dig deeper into this Windows Kernel Protection myself https://t.co/YTUAqGNN7K
分析 MmMapIoSpace 返回 NULL 的内核机制。
内核驱动开发调试参考，对 Windows 内核研究有帮助。
MmMapIoSpace Returns NULL: Tracing the Real Kernel Mechanism Through ntoskrnl https://t.co/dpADGGsGBv
REDSHELL 恶意软件新变种出现在三个 npm 包中。
供应链攻击，npm 包投毒，需检查依赖。
RT @nextronresearch: Follow-up on REDSHELL: we found a newer ELF in three npm packages "hydration-dim-kit", "hydration-dim-ui" and "hydration-ui-dim". It retains the same C2/exfil infrastructure mentioned in our previous tweet but uses a new campaign/beacon ID: a2f65d7abc878ec3c0f2fd1ae67b9db766c3ba357fdfb30fdb1805d07d1ab3bf Just uploaded to VT: https://t.co/MVKuhzkl9V
AnonyMousKIT：AI 驱动的 PhaaS 平台，绕过苹果激活锁。
AI 在攻击中的应用，对移动安全有警示。
RT @socradar: AI is now cold-calling iPhone theft victims. We went inside AnonyMousKIT — an AI-powered PhaaS platform built to bypass Apple's Activation Lock on stolen devices. Thread 🧵 #ThreatIntel #CyberSecurity
钓鱼活动使用 RMM-as-RAT，针对 46 个国家。
钓鱼攻击手法，对防御有参考价值。
https://t.co/URBcFhPXNa analyses a phishing operation delivering an RMM-as-RAT for remote access. The campaign targets 46 countries with lures like US Social Security Administration, PDF documents, invoices, VAT notices & shipping communications. https://t.co/EMbpYL5lXA https://t.co/E7k2tCOhDc
REDSHELL 载荷伪装成 math.mjs，出现在 npm 包中。
供应链攻击持续演化，需加强依赖检查。
RT @nextronresearch: The REDSHELL payload is now disguised as "math.mjs", replacing the previously used .dat and .bin extensions. Our artifact scanner detected it in the npm package "hydration-vli-ui" (v1.0.0), triggering two additional YARA rules designed to flag executables masquerading as script files.
OpenAI 发布 Jalapeño 推理模型，声称速度和效率领先。
新推理模型，可能影响 AI 工具选择。
Jalapeño’s first results show industry-leading speed and efficiency in AI inference - @OpenAI https://t.co/N8cn7jNaQ6
FalconForce 博客：利用 ETW 操纵遥测数据，挑战 EDR 信任。
EDR 规避技术，对红队和防御研究都有价值。
RT @falconforceteam: What if you could leverage Event Tracing for Windows (ETW) to manipulate telemetry data, challenging the trust placed in endpoint detection and response (EDR) tools? 👉 Have a look at our latest blog as presented earlier at Black Hat by @olafhartong: https://t.co/OXo4LxuW22 https://t.co/4MwJ9J3Hvf
Group-IB 分析 Tortoiseshell APT，发现新样本和基础设施。
APT 威胁情报更新，对防御有参考价值。
RT @Cyber_O51NT: Group-IB Threat Intelligence enriched public data on the Tortoiseshell APT, hunted for threats, and found new samples sharing similarities with known malware and additional operational infrastructure. https://t.co/Towb81JK8G
SLEEPWALKER 后门 PoC 发布，并提取独立任务 VM 技术。
新型后门 PoC，对检测和防御研究有价值。
RT @Teach2Breach: PoC is up. Also lifted the task vm portion out into a standalone component technique, in the catalog as Sleepwalker Task VM (BM-T4009). The PoC stays true to the sample, while the isolated task vm technique has been enhanced with BlackMagick bootstrap and opsec policies. https://t.co/hU98FAxLv6
Handle Redirect 技术：重定向内核对象指针到不同 EPROCESS。
内核利用技术，对 Windows 内核安全研究有价值。
RT @Salsa12__: Handle Redirect New Medium post. In this one we will see how to redirect a handle’s kernel object pointer to a different EPROCESS, getting a fully functional handle to a process without any kernel callback ever seeing the operation https://t.co/ut2ieWEcjx https://t.co/Wnzqj7anJM
Canarytokens 上月创建量覆盖全球 75% 国家，发现 AI 云基础设施重大入侵。
蜜罐数据统计，显示 AI 云基础设施攻击趋势。
RT @ThinkstCanary: Last month: 🌍 free https://t.co/WCHU95kI8v were created in ~75% of the countries around the world; 🤯 a single (@ThinkstCanary) customer deployed >500,000 Canarytokens per hour (embedded in their workflows); 💪 we uncovered at least 1 major breach at an AI cloud infra provider https://t.co/s06fGe2YjQ
直播演示用 Frida 破解 Pyarmor 恢复 JavaScript 恶意载荷。
Pyarmor 破解技术，对恶意软件分析有帮助。
RT @InvokeReversing: We've uploaded our live stream from August 18th where we broke Pyarmor with Frida to recover JavaScript malware payloads, enjoy! https://t.co/ZDLi93YUOU
Bitcoin Core Lightning Network 发现严重漏洞，建议节点下线。
加密货币基础设施漏洞，影响相关服务。
RT @pashov: 🚨Seems a Critical Vulnerability was discovered in the Bitcoin Core Lightning Network - take your nodes offline Hidden source code (with the idea of patches not being reverse-engineerable) binaries, holding a fix, will be published in <48hrs Discussion: https://t.co/510e8FzOkJ
用户调侃计算器运行 qwen3.8-27b 接近 Claude。
模型性能讨论，无实质安全内容。
@kyleavery bro my calculator running qwen3.8-27b is basically claude https://t.co/bKhL9kvGTf
OpenCode Go 用户每天花费 1.14 美元，而自建 DGX 成本 1 万美元。
AI 成本对比，对部署决策有参考。
RT @thdxr: the average OpenCode Go user spent $1.14 per day on deepseek flash v4 this past week the dual DGX setup people are running to do the same costs $10,000 it takes 24 years to break even at 10x the usage it takes 2.4 years
用户评论称没有模型能接近 GPT/Claude。
模型性能讨论，无实质安全内容。
@0xTriboulet wake me up when anyone gets even close to gpt/claude
Rust 帮助 CodeDefender 减少 bug，用于 bin2bin 工具。
编程语言选择讨论，对开发有参考。
🦀 Learn how Rust helped CodeDefender iterate successfully with less bugs than C++ while creating some of the tooling for their bin2bin masterpiece. 👉 https://t.co/mFMQbhlrT7 https://t.co/zAHk403sRJ
用户分享链接，无实质内容。
无实质信息。
https://t.co/52wdGWikct
用户质疑 SteamUnlocked 的安全性。
盗版游戏平台风险讨论，无实质安全内容。
Why tf is anyone using SteamUnlocked😭 https://t.co/cOQUMdESEm
发现 StreamYard 被重新签名，疑似恶意软件。
软件签名滥用，对供应链安全有参考。
"Shenzhen Xinfeng E-commerce Co., Ltd." shouldn't be signing other people's software. --- Microsoft fully implemented changes to EV certificates this year: EV certificates no longer give instant trust. As a result, actors who wish to use EV they purchase need to "warm" the cert by getting installs to increase the certificate's reputation. We're seeing this much more frequently. This seems to be a copy of StreamYard which is re-signed: https://t.co/VjIEMzwCB5 We've confidently added it to the Cert Graveyard.
教程：使用 Claude Code 作为操作员创建临时 C2。
AI 用于 C2 的创意，对红队有参考。
How to Create a Makeshift C2 with Claude Code as the Operator - @G3tSyst3m https://t.co/K4RIJCvR4H
Claude Code 配合 Ludus 技能快速搭建 Exchange 测试环境。
AI 辅助搭建测试环境，对红队有参考。
RT @jeffmcjunkin: It's crazy to me that I can kick off a /goal inside Claude Code with the @badsectorlabs Ludus skill and get a fresh, artisanal lab with two Exchange 2019 servers inside for testing https://t.co/hvZ4ZWD7ON Ludus is love. Ludus is life. Use Ludus.
发布论文：Linux ELF 可执行文件的细粒度加载时 ASLR。
Linux 安全机制研究，对防御有参考。
RT @ryan_elfmaster: Pleased to announce the release of my paper in tmp.0ut issue 5: Fine grained load-time ASLR for ELF executables in X86_64 Linux (https://t.co/4knH0geqhS) this issue is absolutely loaded with Gems and Treasures of the underground!
用户手动安装 Arch Linux 到 Framework 13 Pro。
个人技术分享，无安全内容。
RT @rekdt: Took probably 20 minutes to manually setup Arch Linux and kick off the install for Framework 13 Pro https://t.co/TPKMOryqFy
Hugging Face 事件被视为 AI 失控的警告信号。
AI 安全讨论，对理解风险有参考。
RT @tszzl: many people worked incredibly hard on this post and associated report including me whilst everyone took alignment quite seriously before I think no question that this begins a new era. hugging face incident represents reaching a waterline of capabilities that real loss-of-control is possible, and many are taking it as a premonition or ‘warning shot’ of dangers to come. I believe both that alignment is unsolved but also that real progress is possible https://t.co/9oFnONGUuv
发布滥用 Text Template (.tt) 文件的紫队手册。
攻击技术手册，对红队有参考。
Earlier this week, I published the Purple Team playbook for abusing Text Template (.tt) files. If you’re hunting this technique, MSBuild is the loudest of the T4 binaries (Image Load + File Creation). 🔍 𝑫𝒆𝒕𝒆𝒄𝒕𝒊𝒐𝒏 𝑺𝒕𝒓𝒂𝒕𝒆𝒈𝒊𝒆𝒔 ✅️ 𝐏𝐫𝐨𝐜𝐞𝐬𝐬 𝐂𝐫𝐞𝐚𝐭𝐢𝐨𝐧: TextTransform.exe, TextTransformCore.exe, t4.exe & msbuild.exe ✅️ 𝐌𝐨𝐝𝐮𝐥𝐞 𝐋𝐨𝐚𝐝𝐬: 𝑴𝒊𝒄𝒓𝒐𝒔𝒐𝒇𝒕.𝑽𝒊𝒔𝒖𝒂𝒍𝑺𝒕𝒖𝒅𝒊𝒐.𝑻𝒆𝒙𝒕𝑻𝒆𝒎𝒑𝒍𝒂𝒕𝒊𝒏𝒈*.𝒅𝒍𝒍 DLLs ✅️ 𝐅𝐢𝐥𝐞 𝐂𝐫𝐞𝐚𝐭𝐢𝐨𝐧: T4 and output files under 📁 𝑪:\𝑼𝒔𝒆𝒓𝒔\<𝒖𝒔𝒆𝒓>𝑨𝒑𝒑𝑫𝒂𝒕𝒂\𝑳𝒐𝒄𝒂𝒍\𝑻𝒆𝒎𝒑\ https://t.co/PHnhxu8HC5
静态反虚拟化 Tencent VM 的研究。
逆向工程技术，对恶意软件分析有帮助。
Static Devirtualization of Tencent VM (@BackEngineerLab) https://t.co/pUNxbrtbGk #infosec https://t.co/BLiYLl2cKq
游戏黑客技术对安全研究有巨大价值。
低层知识讨论，对安全研究有参考。
RT @cr3ghost: If you work in infosec and still dismiss game hacking as 'just cheats', you're leaving a ridiculous amount of low-level knowledge on the table. Guided Hacking has 280+ chapters covering Windows internals, x86/x64, PE internals, IDA, WinDbg, process injection, shellcode, kernel drivers, BYOVD, hypervisors, anti-debugging, EDR/evasion concepts, exploit development, fuzzing, reverse engineering and anti-cheat internals. There's an entire Game Hacking Bible, Game Hacking Fundamentals material, plus Anti-Cheat Development and Devirtualization courses on the way. For a relatively small cost compared with many infosec courses, certifications and training programs, the amount of transferable knowledge is kind of insane. 1. Reverse engineering. 2. Malware analysis. 3. Detection engineering. 4. Vulnerability research. 5. Exploit development. 6. Software engineering. 7. Windows kernel. The game is different. The primitives are the same. The objective changes. Courses: https://t.co/e8lBjuAdce Videos: https://t.co/1Nc9krp7Dw Awesome work @GuidedHacking aka Rake for consistently putting out quality content and making deep technical concepts more approachable. You've done a great job building a genuinely valuable learning platform and giving people a practical path to develop serious low-level skills. #ReverseEngineering #WindowsInternals #Infosec
GLM-5.3-Flash 漏洞发现能力比 GPT-5.6-Luna 高 15%。
模型安全能力对比，对选择 AI 工具有参考。
RT @pilvar222: GLM-5.3-Flash (Ox Alpha) came out of stealth today, and it turns out its pricing is amazing 🔥 - It finds 15% more vulnerabilities than GPT-5.6-Luna at pass@1 and matches it at pass@3 - The interesting part: its discounted pricing makes it Pareto frontier on both pass@1 and pass@3 GLM-5.3-Flash is the smallest frontier model in cyber 1/3 🧵
Firecracker 沙箱未被逃逸，增强信心。
沙箱安全性讨论，对云安全有参考。
RT @InsanityBit: Surely at this point there has been >$1M tokens spent on trying to escape Firecracker and it hasn't happened. This should give everyone confidence that the "vulnpocalypse" is solved largely by doing the obvious work.
红队成员想象有人用漏洞做了很酷的事。
无实质内容。
RT @HackingLZ: As a red teamer, I always like to imagine someone did something really cool with a bug like this before reporting it. https://t.co/mKVmdF3gp1
NetAskari 分析泄露的 ZRON 数据。
威胁情报分析，对了解攻击基础设施有参考。
Read up our analysis of the leaked ZRON data: https://t.co/CaKpdd7BVw https://t.co/tIpYdSVVeV
视频演示使用 MSBuild.exe 执行 .tt 文件中的代码。
攻击技术演示，对红队有参考。
RT @ipurple: A quick video of how to use MSBuild.exe to target a .csproj file that has been tampered with multiple text template files and execute code embedded in .tt files.
交互模式使查找更容易。
工具使用技巧，无实质安全内容。
RT @irsdl: The interactive mode makes things easier if you are just looking for stuff. https://t.co/0OAMizOSKL
Outflank 介绍外部开发者如何将工具集成到 OST。
工具集成流程，对红队工具有参考。
RT @OutflankNL: Ever wondered how an external developer can work with Outflank to get their tool into OST? We explore the process of incorporating infraRED, which can be used for auto deployment of RT infra (c2, redir, multi cloud, domains, email).Big thanks to @dsec_net! https://t.co/MuqjqlExom https://t.co/wOKBpwyPhr
AI 导致恶意软件发现率指数级增加，难以判断持久性。
AI 对恶意软件分析的影响，对防御有参考。
RT @Myrtus0x0: Anyone else feel like their rate of malware discovery has exponentially increased since AI? its begun to be a problem where it takes extra time to determine if new malware is going to be around long enough to dedicate human hours to it
评估模型是否仅从源代码找到漏洞，需要 PoC 在真实站点触发。
AI 漏洞评估标准讨论，对安全评估有参考。
And so when we ask "can models find things from source code alone" in an eval we default to the standards of evidence we'd require during our normal assessments: can the PoC trigger on a live site?
XBOW 优先报告动态验证的漏洞。
AI 漏洞挖掘方法讨论，对安全研究有参考。
It's a confusing graph but IMO it does reveal an interesting difference in philosophy wrt how XBOW is built – from the beginning we *strongly* prioritized only reporting findings we could prove were real dynamically (see my BH 2025 talk for examples)!
YSoNet v2026.8.1 发布，新增多个 .NET 反序列化 gadget。
.NET 反序列化工具更新，对漏洞利用有帮助。
RT @irsdl: YSoNet v2026.8.1 is live. 50 to 62 gadgets, new .NET Framework 2.0 to 3.5 support, dedicated 4.0 coverage, stronger testing, and 508 archived references for researchers in the markdown format! 🔥 Thanks @cjm00n and @sinsinology! https://t.co/JgGRh1hbpG
发现 StreamYard 被重新签名，疑似恶意软件。
软件签名滥用，对供应链安全有参考。
"Shenzhen Xinfeng E-commerce Co., Ltd." shouldn't be signing other people's software. --- Microsoft fully implemented changes to EV certificates this year: EV certificates no longer give instant trust. As a result, actors who wish to use EV they purchase need to "warm" the cert by getting installs to increase the certificate's reputation. We're seeing this much more frequently. This seems to be a copy of StreamYard which is re-signed: https://t.co/VjIEMzwCB5 We've confidently added it to the Cert Graveyard.
教程：使用 Claude Code 作为操作员创建临时 C2。
AI 用于 C2 的创意，对红队有参考。
How to Create a Makeshift C2 with Claude Code as the Operator - @G3tSyst3m https://t.co/K4RIJCvR4H
Claude Code 配合 Ludus 技能快速搭建 Exchange 测试环境。
AI 辅助搭建测试环境，对红队有参考。
RT @jeffmcjunkin: It's crazy to me that I can kick off a /goal inside Claude Code with the @badsectorlabs Ludus skill and get a fresh, artisanal lab with two Exchange 2019 servers inside for testing https://t.co/hvZ4ZWD7ON Ludus is love. Ludus is life. Use Ludus.
发布论文：Linux ELF 可执行文件的细粒度加载时 ASLR。
Linux 安全机制研究，对防御有参考。
RT @ryan_elfmaster: Pleased to announce the release of my paper in tmp.0ut issue 5: Fine grained load-time ASLR for ELF executables in X86_64 Linux (https://t.co/4knH0geqhS) this issue is absolutely loaded with Gems and Treasures of the underground!
Hugging Face 事件被视为 AI 失控的警告信号。
AI 安全讨论，对理解风险有参考。
RT @tszzl: many people worked incredibly hard on this post and associated report including me whilst everyone took alignment quite seriously before I think no question that this begins a new era. hugging face incident represents reaching a waterline of capabilities that real loss-of-control is possible, and many are taking it as a premonition or ‘warning shot’ of dangers to come. I believe both that alignment is unsolved but also that real progress is possible https://t.co/9oFnONGUuv
发布滥用 Text Template (.tt) 文件的紫队手册。
攻击技术手册，对红队有参考。
Earlier this week, I published the Purple Team playbook for abusing Text Template (.tt) files. If you’re hunting this technique, MSBuild is the loudest of the T4 binaries (Image Load + File Creation). 🔍 𝑫𝒆𝒕𝒆𝒄𝒕𝒊𝒐𝒏 𝑺𝒕𝒓𝒂𝒕𝒆𝒈𝒊𝒆𝒔 ✅️ 𝐏𝐫𝐨𝐜𝐞𝐬𝐬 𝐂𝐫𝐞𝐚𝐭𝐢𝐨𝐧: TextTransform.exe, TextTransformCore.exe, t4.exe & msbuild.exe ✅️ 𝐌𝐨𝐝𝐮𝐥𝐞 𝐋𝐨𝐚𝐝𝐬: 𝑴𝒊𝒄𝒓𝒐𝒔𝒐𝒇𝒕.𝑽𝒊𝒔𝒖𝒂𝒍𝑺𝒕𝒖𝒅𝒊𝒐.𝑻𝒆𝒙𝒕𝑻𝒆𝒎𝒑𝒍𝒂𝒕𝒊𝒏𝒈*.𝒅𝒍𝒍 DLLs ✅️ 𝐅𝐢𝐥𝐞 𝐂𝐫𝐞𝐚𝐭𝐢𝐨𝐧: T4 and output files under 📁 𝑪:\𝑼𝒔𝒆𝒓𝒔\<𝒖𝒔𝒆𝒓>𝑨𝒑𝒑𝑫𝒂𝒕𝒂\𝑳𝒐𝒄𝒂𝒍\𝑻𝒆𝒎𝒑\ https://t.co/PHnhxu8HC5
静态反虚拟化 Tencent VM 的研究。
逆向工程技术，对恶意软件分析有帮助。
Static Devirtualization of Tencent VM (@BackEngineerLab) https://t.co/pUNxbrtbGk #infosec https://t.co/BLiYLl2cKq
GLM-5.3-Flash 漏洞发现能力比 GPT-5.6-Luna 高 15%。
模型安全能力对比，对选择 AI 工具有参考。
RT @pilvar222: GLM-5.3-Flash (Ox Alpha) came out of stealth today, and it turns out its pricing is amazing 🔥 - It finds 15% more vulnerabilities than GPT-5.6-Luna at pass@1 and matches it at pass@3 - The interesting part: its discounted pricing makes it Pareto frontier on both pass@1 and pass@3 GLM-5.3-Flash is the smallest frontier model in cyber 1/3 🧵
Firecracker 沙箱未被逃逸，增强信心。
沙箱安全性讨论，对云安全有参考。
RT @InsanityBit: Surely at this point there has been >$1M tokens spent on trying to escape Firecracker and it hasn't happened. This should give everyone confidence that the "vulnpocalypse" is solved largely by doing the obvious work.
NetAskari 分析泄露的 ZRON 数据。
威胁情报分析，对了解攻击基础设施有参考。
Read up our analysis of the leaked ZRON data: https://t.co/CaKpdd7BVw https://t.co/tIpYdSVVeV
视频演示使用 MSBuild.exe 执行 .tt 文件中的代码。
攻击技术演示，对红队有参考。
RT @ipurple: A quick video of how to use MSBuild.exe to target a .csproj file that has been tampered with multiple text template files and execute code embedded in .tt files.
Outflank 介绍外部开发者如何将工具集成到 OST。
工具集成流程，对红队工具有参考。
RT @OutflankNL: Ever wondered how an external developer can work with Outflank to get their tool into OST? We explore the process of incorporating infraRED, which can be used for auto deployment of RT infra (c2, redir, multi cloud, domains, email).Big thanks to @dsec_net! https://t.co/MuqjqlExom https://t.co/wOKBpwyPhr
AI 导致恶意软件发现率指数级增加，难以判断持久性。
AI 对恶意软件分析的影响，对防御有参考。
RT @Myrtus0x0: Anyone else feel like their rate of malware discovery has exponentially increased since AI? its begun to be a problem where it takes extra time to determine if new malware is going to be around long enough to dedicate human hours to it
评估模型是否仅从源代码找到漏洞，需要 PoC 在真实站点触发。
AI 漏洞评估标准讨论，对安全评估有参考。
And so when we ask "can models find things from source code alone" in an eval we default to the standards of evidence we'd require during our normal assessments: can the PoC trigger on a live site?
XBOW 优先报告动态验证的漏洞。
AI 漏洞挖掘方法讨论，对安全研究有参考。
It's a confusing graph but IMO it does reveal an interesting difference in philosophy wrt how XBOW is built – from the beginning we *strongly* prioritized only reporting findings we could prove were real dynamically (see my BH 2025 talk for examples)!
YSoNet v2026.8.1 发布，新增多个 .NET 反序列化 gadget。
.NET 反序列化工具更新，对漏洞利用有帮助。
RT @irsdl: YSoNet v2026.8.1 is live. 50 to 62 gadgets, new .NET Framework 2.0 to 3.5 support, dedicated 4.0 coverage, stronger testing, and 508 archived references for researchers in the markdown format! 🔥 Thanks @cjm00n and @sinsinology! https://t.co/JgGRh1hbpG