⚡Megatron LLM Hub
全部
Twitter

推特安全流

头条6

默认配置即受影响,无需登录即可触发,自建服务与客户端双向暴露

RT @unusual_whales: BREAKING: OpenAI says it found a critical flaw in Brotli, a widely used compression library built into software like Node.js, Bun and Envoy. The bug could let an attacker take over a system just by sending it malicious compressed data, with no login needed. That works both ways: a hacker could hijack an unpatched server, or a malicious server could hijack an unpatched user’s device.

rce供应链nodejs @0xocdsec 原文 ↗

PoC 公开且默认部署可被远程执行,mod_php 与 FPM 的差异直接决定暴露面

RT @JohnHultquist: A brief update on the critical vulnerabilities in PHP:  We have not yet validated exploitation in the wild and we have no evidence of successful wide scale exploitation. There are currently several exploit PoCs publicly circulating, many of which appear to be "vibe-coded" and are incapable of remote code execution. We have validated some PoCs, however, that are capable of RCE against default deployments of mod_php but not PHP-FPM. Nonetheless, exploitation may have already begun in a limited fashion and the threat is viable and very serious. Defenders should: ​1) Patch & Restart by upgrading to PHP 8.5.12, 8.4.27, 8.3.36, or 8.2.35 and restarting web services ​2) Layer Temporary Safeguards at the network edge (WAF/proxy rules) and server configuration level if immediate patching requires a maintenance window ​3) Protect Credentials by monitoring for abnormal session traffic and rotating keys and passwords on compromised internet-facing PHP systems exposed prior to patching 🔗 We are continually updating our Cloud Security blog post on this topic with real-world guidance and actionable defense strategies as we learn more. Please keep monitoring these posts for updates as you prioritize your patching efforts: https://t.co/uDnd6gk988  🔗 Technical details from Wiz: https://t.co/RqmTqEDM2V 🔗 Details of the exposure from VirusTotal: https://t.co/Toix3AM0yq

rcephppoc @cyb3rops 原文 ↗

缓解只需改一个 ini 开关,仅运行时切换 serialize_handler 到 php 的非默认应用受影响。

RT @ThePHPF: Set session.upload_progress.enabled=0 till next PHP release to block an unlikely RCE vector (CVE-2026-115560). Already recommended for CVE-2026-109619 Wed, keep disabled. Only apps with non-default serialize_handler switching it to "php" at runtime can be affected. #php #phpc

rcecvephp缓解 @artem_i_baranov 原文 ↗

未授权远程代码执行,所有对外 PHP 服务均受影响,补丁已发布可立即升级

🚨 PHP2Shell: two critical PHP vulnerabilities are now patched. PHP has fixed CVE-2026-109618 and CVE-2026-109619 (CVSS 9.2). Both can let an unauthenticated attacker run code on the server. If you run PHP: update every runtime serving web traffic (check the version behind PHP-FPM or Apache, don’t rely on the CLI version alone), restart PHP workers, and review activity from at least 17 October onwards. Full breakdown: https://t.co/i1iKRJ0O5S

rcecvephp @HacktronAI 原文 ↗
必看24

给出漏洞发现路径与受影响组件清单,可据此排补丁优先级

RT @itszn13: The critical Brotli RCE vulnerability discovery came out of our large scale infrastructure code scanning efforts Great research by @k_tr4n and @_sy1vi3 in assessing the widespread impact of this bug Update your Node.js, Bun, Envoy, anything using Brotli ASAP! https://t.co/c12Kl59jr9

rce供应链nodejs @lcfr_hax 原文 ↗

极热门基础依赖,补丁已发布但下游包面广,影响范围仍在扩散。

RT @AikidoSecurity: ⚠️ Python packages are vulnerable to remote code execution via the same node2shell CVE-2026-111497. You may be affected if you are using the Brotli library. Note this is a very popular dependency (⬇️100m/month) Other major services likely affected. We will communicate ASAP when we are able to. Aikido detects this. Patches available.

供应链rcecvepython @cyb3rops 原文 ↗

签名强制并非终点,受影响主机含域控,缓解措施尚未覆盖。

RT @cr3ghost: SMB signing enforced? You might still be vulnerable to relay. Really interesting research from @aas_s3curity demonstrating how SMB2/3 re-authentication can bypass enforced signing on affected Windows systems, potentially including Domain Controllers. The technique: • Establish an SMB session using a low-privileged account • Relay privileged NTLM or Kerberos authentication • Re-authenticate the existing session as that identity • Abuse retention of the original signing key • Access the privileged session while signing with a key you control The underlying issue is a missing identity check during re-authentication. The author released reauthx, a krbrelayx fork, alongside tooling to identify affected hosts. Important: September 2026 Windows updates appear to address this. Successful exploitation still requires a suitable privileged authentication and other relay preconditions. Excellent research building on work from Dirk-jan Mollema, Synacktiv, SySS, Semperis and SpecterOps. Write-up: https://t.co/ItQDtR86E1 PoC: https://t.co/J0YLr6uxpX #ActiveDirectory #RedTeam #WindowsSecurity

smbntlm_relay域控绕过 @cr3ghost 原文 ↗

PHP 部署面极广,漏洞未修补且已进入公开讨论。

RT @Dinosn: URGENT - PHP Warns of Another Unpatched RCE Two Days After Emergency PHP2Shell Patches https://t.co/Gv88QF8tWc

rcephpcve @artem_i_baranov 原文 ↗

被当作已缓解的域控场景仍可被 relay,需立即核对配置。

RT @SpecterOps: New research from @GrayHatKiller: NTLM relay to domain controllers with LDAP signing enforced. Patch your DCs. 👇 https://t.co/jNcQOmH72U

ntlm_relay域控ldap提权 @0xocdsec 原文 ↗

本地 AI 客户端会话与开发凭据已成为窃密木马的定向收集目标。

Infostealers are hijacking the AI train. What’s Aokigahara after? 🎯AI sessions from Claude and ChatGPT Desktop 🔑IDE data and developer secrets 💰Browser credentials and crypto wallets Your AI workspace is on the collection list. Read more 👇 https://t.co/yOJYkm83UO

ai_agentllm窃密报告 @_CPResearch_ 原文 ↗

agent 辅助代码审计已产出真实预认证 RCE,审计速度差需重新评估。

RT @lorenc_dan: The actual prompt for this was "let's look for pre-auth RCEs in PHP". I'm sharing this so people know this stuff is real and to take agent-assisted scanning seriously. I kicked if off using Claude Mythos 5.1 (in stock Claude Code) but it uses a skill to drive a harness we built over the last few months. I don't know if the harness is amazing, but it certainly has a scary track record so far. It's setup to do an agent guided tree search across a codebase, and runs in microvm sandboxes that all have the full source code for the target, as well as the source for any transitive dependencies. When I built the harness I basically just fed Claude a bunch of Project Zero blog posts describing how they did research and had it build a harness that ran like that, but on our tech stack. The first phase is just setup and can run with any model, it just does that fetching and assembles the kit environment, and gets a target up and running alongside the source code. Then we just do some basic suspend/resume stuff over to the hunt phase, where one agent supervises the tree, and the first node kicks off. Each node is context-aware and aware of the work of all the others. I think I ran this with parallelism 4? So 4 running at any one time. The supervisor can shape the tree and boost nodes or prune. Any node can add more below it with targeted instructions. A typical run looks like the first node doing a codebase review or building and running a fuzzer or something, all depending on the target. Then based on whatever it finds, it'll queue 4-5 children that go look into those areas. The main system guidance is to validate everything 3 times in clean VMs, and we typically require full exploit chains/POCs. Here's what this run looked like. We're doing this with a bunch of other open source projects now and companies. We're doing a lot of this for free, even with private companies, across critical infrastructure. We're directly partnered with Anthropic, and can do this even if you don't have direct access to a frontier model yourself.

ai_agentllmrce审计 @artem_i_baranov 原文 ↗

自托管 Agent 沙箱的隔离假设,正被形式化验证重新审视。

hear me out, mathematicians entering ai alignment maybe consider this challenge. i think its probably millennium level problem(?) take any complex real world software, like a browser or a kernel. let's pick a kvm based sandbox, which i suspect is somewhat less complex than chromium(?), running an increasingly capable, potentially adversarial AI agent. we want to formally prove one simple property, no matter what code the agent executes inside the VM, it cannot escape and compromise the host. for that we need to prove the entire relevant software stack, kvm, cpu virtualization, kernel, virutal devices that could affect the isolation. (to make things easiers, lets assume hardware wont have weird states) can we mathematically establish whether formally verifying a system of this complexity is even computationally feasible?

ai_agent逃逸议题 @S1r1u5_ 原文 ↗

漏洞发现从人工审计转向规模化证明,防守成本结构随之改变。

RT @AnthropicAI: New on our Frontier Red Team blog: using Claude to formally verify software at large scale. Formal verification can prove that code is free of certain bugs, but historically it has only been feasible for small codebases (or required a lot of expert time). With AI models, however, this is beginning to change. In collaboration with @theoremlabs, we used Claude to prove security properties in over one hundred real-world open-source libraries over the span of a few weeks. Read our research in full here: https://t.co/zUyx8CjBEk

llm报告ai_agent @FuzzySec 原文 ↗

高关注度 PHP RCE 的真实可利用性判断,直接影响排查优先级。

RT @MalwareTechBlog: Current information I have on php2Shell: - The RCEs requires the attacker to obtain memory addresses which are typically protected by ASLR. - I suspect the original author (lcfr_hax) likely has a reliable exploit which leads to stable code execution. - Most of the PoCs being posted are completely non-functional AI slop. - There is at least 1 working non-slop exploit being passed around, but it's reliant on bruteforcing for ASLR bypass, so not particularly impactful. - Rumor of ITW exploitation are patchy. There are attempts being seen using the non-functional leaked RCE, or variants that are config / system specific. - Some rumors point towards a known-working exploit being used by threat actors (I've personally confirmed this specific exploit works, but the ASLR bypass technique is unreliable at best and massively limits the potential impact). - I'd expect heavy ITW exploitation in the next few weeks based on how many people are looking at this. Though, success rates will vary heavily based on the specific exploit chain.

rcepocllm @artem_i_baranov 原文 ↗

大量二手报道仍引用“已被利用”,据此决策会误判暴露面的优先级。

RT @ImposeCost: "The previous update (October 22 16:18 UTC) reported that Sophos observed successful exploitation of PHP2Shell in customer environments. Upon further analysis, we are correcting that assessment as Sophos customer telemetry shows scanning and a small number of failed exploitation attempts. As of this update, there is no evidence of a successful compromise due to these vulnerabilities." https://t.co/BC5caSSSiR

php事件cve @domchell 原文 ↗

显示低成本模型已能批量复现真实 CVE,本地部署漏洞挖掘的门槛随之下降。

RT @pilvar222: Unsurprisingly, DeepSeek V4.1 Flash 1017 is mogging all other models on our cybersecurity benchmark - At pass@3, it rediscovered 87.5% of the benchmark CVEs (28/32). That ties DeepSeek V4 Pro 0813 for 2nd place, behind only GPT-6 Astra (90.6%) - Single-run recall is 70.8%, matching Opus 5 and Grok 4.7. The previous Flash was at 65.6% - Precision went from 78.9% to 86.4%, so most false positives are now gone Cost: $2.40 per CVE found. That’s 27x cheaper than Opus 5 ($64.96) and 56x cheaper than Astra ($135.23) @deepseek_ai keeps pushing the Pareto frontier 1/4 🧵

llmai_agentpoc报告 @cyb3rops 原文 ↗

环境细节不再构成门槛,利用开发周期被压缩到天级。

@LowLevelTweets @MalwareTechBlog @HackingLZ its just reduces the speed of exploitation, every single exploit we sprayed on target was improvised here not knowing heap allocator, debian image, even libheif version. https://t.co/amYfP3x0NF every single target was compromised in a day or two

llmexploitpoc @S1r1u5_ 原文 ↗

固件锁死仍靠 0day 拿到 root,暴露消费级设备默认开启的数据采集面。

RT @0x0SojalSec: WTF, $3000 Samsung TV has an always-on mic, automatic content recognition, geo tracking, network mapping, and ships your data to Samsung + 400 partners. Bro Rooted a brand-new 2025 Samsung TV after noticing absurdly targeted ads. Firmware was locked down, so zero-days it was. Got a root shell. then found: always-on voice service, ACR, geo, full telemetry pipeline, and four separate DRM systems. This isn’t a bug, It’s the product bro.

固件0day隐私 @0xocdsec 原文 ↗

静态链接副本常被漏查,TLS 证书压缩是尚未评估的攻击面

RT @julianor: Brotli CVE-2026-111497 goes well beyond Node.js. Quick search shows unpatched Brotli decoders in Java HTTP stacks, WOFF2 libs, CF workerd. TLS certificate compression (RFC 8879) is another attack surface worth investigating. Check statically linked Brotli copies.

cvetls供应链 @0xocdsec 原文 ↗
推荐15

原生二进制、无 PowerShell 与 .NET 依赖,受限环境也能直接跑常见提权检查

RT @0xby7e_me: Just open-sourced WinLPEAuditor - a Windows LPE misconfiguration auditor built in C++. Checks for unquoted service paths, AlwaysInstallElevated, dangerous token privileges, modifiable service binaries, and more. Native binary, no PowerShell, no .NET. https://t.co/4eNt4kJmWQ

lpe工具windows @ipurple 原文 ↗

免费开放的板级走线工具,硬件取证与改机分析可直接上手。

RT @pascal_gujer: 🔍 One loupe to find the chip. Another to follow the signal. Meet BoardLoupe — our new open-source boardview tool for hardware hackers and reverse engineers! Already being put to work by students at our Breaking BitLocker training in Zurich. 🔬💻 🔗 https://t.co/9yxBsxNdMU #HardwareHacking #OpenSource #DigitalForensics

工具硬件逆向 @SkelSec 原文 ↗

汇总已公开的绕过手法,评估 Windows 应用控制策略时可直接对照。

RT @sekurlsa_pw: Reposting because it’s good (you might have seen this before but it gets updated) ”UltimateWDACBypassListPublic A centralized resource for previously documented WDAC bypass techniques” Author: @bohops

绕过windows工具 @bohops 原文 ↗

利用视觉通道绕过 LLM 内容过滤,可直接复现验证既有防护。

I built a tool that uses my previous LLM image packing research and lets you turn text into images that bypasses filters of many model providers, especially Gemini: https://t.co/dyVSpBtu0e https://t.co/aBydvBY0WN

llm绕过工具多模态 @in19h 原文 ↗

展示了蜜罐采集加 agent 自动打标的可复用架构,适用于 PHP RCE 在野监测。

RT @0x4D31: monitoring the php2shell situation with my 🍯phPot and @perplexity_ai Computer! running eBPF-monitored PHP containers across multiple app personas behind a Finch proxy to capture TLS/HTTP fingerprints alongside web, app, and kernel-level signals. a sealer packs and pushes each window to a datastore for a pplx Computer agent to analyze, populate the dashboard, and flag the interesting ones!

ai_agent蜜罐php工具 @_xpn_ 原文 ↗

签名开发工具被滥用为执行链,可作为 LOLBAS 类检测项。

While looking for MSBuild alternatives, I came across VisualStudio's TextTransform utility which can execute C# code embedded in `.tt` files - something I haven't seen in the wild before, so naturally, I had to make a tool around it to run shellcode ;) https://t.co/98CD8P8YuR https://t.co/iWI5KtaFZ5

工具绕过rce @whokilleddb 原文 ↗

给出了自建 LLM 扫描管道的分层结构与成本,可直接照搬到自托管环境。

Our OSS artifact scanning picked this up too The funny part, at least for us, is that this is a side project built around the scanner and detection rules we already ship for endpoints. Its initial LLM triage costs about $3 a day. We download packages from public registries, unpack them, and run their files through THOR Thunderstorm. The generic detection rules flag suspicious code, obfuscation and recurring attacker tradecraft. Only artifacts with findings reach the LLM, which gets the matched strings and surrounding context to assess what’s going on. Across roughly 114,000 artifacts a day, about 4% reach that initial LLM pass. The rest never generate a token bill. In this case, an existing generic Python rule flagged the malicious https://t.co/3A11Y4XWdW. We already had the detection logic; we’re just applying it to packages downloaded from registries as well as files found on endpoints. And that’s what amuses me about this. Supply-chain security attracts substantial funding, and specialist vendors are reporting this compromise. We picked it up by reusing years of endpoint detection engineering in an internal project, with an initial LLM triage bill averaging $2.88/day. Pretty useful bycatch. We described the pipeline here: https://t.co/bBDiBnUVti

供应链llm工具检测规则 @cyb3rops 原文 ↗

现成检测内容可直接部署,省去从原始日志自建规则的环节

We just released a new Splunk ESCU version 6.9.1 with initial coverage for PHP2Shell - Grab the update today https://t.co/HP0RzuWKjc Check out the dedicated analytic story for more details - https://t.co/E7fhGqcFJV Big shoutout to Onur Mustafa Erdogan @raven_tait @SnekCharmerr for making this happen

检测php规则 @nas_bench 原文 ↗

真实利用流量与规则可直接用于自查是否已被入侵

RT @Kostastsale: Really proud of what we managed to capture and put together at Threat Hunting Labs around PHP2Shell in just a day of non-stop work. I'm tired but wanted to share this quickly with the community. We took real exploitation activity and turned it into something the whole community can benefit from. Raw logs, hunting queries, Sigma and Suricata rules, a PCAP, and a full investigation with our own analysis. We normally reserve the detailed researcher analysis for subscribers, but we've made everything free for this one. No gatekeeping here, this is free for everyone. Hope you find it useful! https://t.co/CVfthEAEVZ

检测pcapphp @nas_bench 原文 ↗
更多95

模型开放策略表态,对本地部署选型有间接参考

RT @rohanpaul_ai: Sam Altman says OpenAI's current best internal model is too powerful to release openly any time soon. Before that could happen, the whole cyber defense setup of the internet would need to change and the world would need a new layer of protection against biological threats. ---- Full video on "Stanford Graduate School of Business" YouTube channel, (link in comment)

llmai_agent @0xocdsec 原文 ↗

新机型上市即出现可用越狱,说明攻击面在新硬件上被迅速摸清。

RT @MacRumors: iPhone Duo Jailbroken on Launch Day https://t.co/3gpolB6AOv https://t.co/iKBksEzuNY

移动端越狱ios @0xocdsec 原文 ↗

从监控组件漏洞延伸到大规模设备指纹采集,暴露 GPU 管理面的信息泄露。

RT @YakirKad: We found NVIDIA’s latest Blackwell Ultra B300 GPUs on Chinese networks. 👀🇨🇳 Together with @JordanNanos from @SemiAnalysis_, we analyzed data from 12,000+ NVIDIA GPUs. It all started with a vulnerability in NVIDIA’s GPU monitoring software. Here’s how we uncovered it. 🧵👇

漏洞gpu报告 @0xocdsec 原文 ↗

身份联合登录是钓鱼主通道,动态弹窗显著提高可信度与成功率。

I'm currently working on expanding the Browser-in-the-Browser (BITB) phishing technique to make it more versatile by allowing spoofed popup windows to be spawned dynamically on phishing pages. This will allow to reliably perform phishing attacks against websites that use identity federation, which is just a fancy term for "signing in with your Google or Microsoft account". When the user picks an option to sign in with their Google or Microsoft account, Evilginx Pro spawns a BITB popup window with a phishing page generated by a different phishlet. That way, the attacker can phish a different account, which, once taken over, will allow access to the one initially targeted. In the demo below, you can see how the Google and Microsoft sign-in phishing pages are loaded within BITB popup windows. Only the Apple sign-in option spawns the legitimate browser popup window with a legitimate sign-in page. Google => PHISHING ❌ Apple => REAL✅ Microsoft => PHISHING ❌ Could you spot the fake popups? P.S. Coming soon to Evilginx Pro 5.0 🥳

钓鱼bitb恶意工具 @mrgretzky 原文 ↗

给出 LLM 提升 fuzzing 覆盖率的具体工程路径与权衡取舍。

RT @OffensiveAIcon: New talk up OAIC 2026: Vibe Fuzzing with Sam Lovejoy @_dinolover38 of IBM X-Force Offensive Research. Sam explores how graph-based inputs, program state, and LLM guidance come together in an AI-guided fuzzing engine- and the tradeoffs behind it. Watch the full talk: https://t.co/zFcLR5AvZd #OAIC #OffensiveAI #Fuzzing

llmfuzzing议题 @chompie1337 原文 ↗

开放权重意味着可自托管,是本地 agent 部署的候选底座。

RT @Kimi_Moonshot: Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model. 🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal. 🔹 Full-project coding: games in Unreal, Unity & Godot. 🔹 Excels at parametric 3D modeling in Blender & CAD. 🔹 More reliable at finance work: Accurate modeling with hallucination down by 65%. 🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks. Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on https://t.co/85jA71X9gw, Kimi Work, Kimi Code, and the Kimi API. Open weights by Nov 4, 2026. 🔗 API: https://t.co/XCrgjXAqMw 🔗 Tech blog: https://t.co/5iPP0MBej6

llm开源权重自托管 @cr3ghost 原文 ↗

MCP 工具生态扩张同时放大 agent 能力与可用攻击面。

mcp-gamehacking — MCP server exposing a 4,100+ skill game-hacking and reverse-engineering knowledge base with source archives, specialized agents, subagents, plugins and workflow tools for Claude, Cursor and other MCP clients. https://t.co/AosLiGjqDy https://t.co/fk8gESKcYc

ai_agentmcp工具逆向 @KitPloit 原文 ↗

在编译后阶段验证可减少源码语义假设,关系 AI 生成代码可信性。

RT @hillelogram: It is a weird weird year to be a formal methods expert. Some unorganized thoughts on the new @AnthropicAI and @theoremlabs post on formal verification (https://t.co/YoymKXYFJa): - This is very cool. - The main innovation is verifying a later stage of the program. Normally provers take the source code, LLM it into a theorem language, and then prove that, which means you have to make assumptions about source semantics and compilation properties. It's very easy to make a mistake in translating and prove the wrong thing! What they instead do is compile the code into assembly, and then (if I understand this correctly), compile that assembly into a theorem language (which they call LPS). So as long as the compiler is faithful you can skip the whole divergence problem. - I'm glad they clarify that you still have to trust the translation process, the "hardware, OS, OCaml" compiler, the LPS model, and also assume that the program is single threaded with fixed memory. The single biggest source of errors in formally verified code is when the proof assumes something that isn't true in practice. It's really, really important to state your assumptions! - I'm not clear why they need to compile the fuzzer into the theorem. Is it to act as a description of valid inputs and workload? Maybe to embed assertions in as theorems? - Holy crap, 180 million lines of Rocq. Is that more Rocq than exists in the world already? I notice they say a 600,000 line proof corresponds to roughly 20k CPU instructions. That would be like, 1-2k lines of source code at most? Really goes to show how inefficient this approach is and how it only works because you can throw LLMs at the problem. - This, probably deliberately, doesn't address the single biggest gap in AI-based verification: AIs cannot come up with good properties. Doesn't matter if you're doing unit testing, property testing, formal verification. I always assumed this was a "frontier labs haven't trained for it yet" thing, but if Anthropic hasn't talked about that here, maybe they've tried and haven't been successful yet. That's interesting to me. Instead, they take a known universal property (this code doesn't crash or get flagged by UBsan) and prove that instead. Still very useful, but makes it harder to generalize to proving arbitrary things about arbitrary programs. - Anybody reading the official post should also read their expert review appendix (https://t.co/qQU0E0nJwW). It points out some interesting things, like the fact that most of the code the LLM wrote wasn't used in the proof (over one billion lines of Rocq), that they had to prove the equivalence between fuzzed and unfuzzed programs, and (this is most interesting), some experiments find that the LLM-method of proofwriting may not scale to more complex things. - Also read the description of LPS (https://t.co/hRTCL05X0H) to understand the limitations of the assembly -> Rocq compilation. In particular, it does not handle concurrency at all (which is one of the main use cases for formal methods!) or OS signals. Also it can't prove things about program arguments with slash characters, meaning they can prove stuff about `rm foo` but not `rm foo/bar`. - The official skill (https://t.co/weDOdmJUfO) doesn't tell us much about the nature of FM and AI. Probably too much for me to hope of any skill. - Some of the article's specific caveats and throat clearings make me think someone at Anthropic reads my stuff ( ͡° ͜ʖ ͡°)

llm形式化验证报告 @0xocdsec 原文 ↗

RT @MTSlive: SITUATION DETECTED: CloudFlare CEO Matthew Prince says he is aware of additional severe vulnerabilities in other pieces of critical software, and said “I think we’re going to have a couple years where every single week it’s going to feel like the sky is falling,” per Reuters.

cve供应链 @0xocdsec 原文 ↗

Two identical Youtube uploads of the same video twice. One of them (2) is discoverable everywhere, whereas video 1 is only searchable in Chinese. The only thing that differ are the chosen keywords, but not by much. Which keyword is causing the shadow ban for video 1? JB #seo #creator

平台审核 @drJonasBirch 原文 ↗

I strongly believe a part of why so many critical vulnerabilities are happening all over is the proliferation of this mental model from the frontier AI companies. If researchers could reliably access frontier capability for offensive and defensive work, we wouldn’t be on equal footing with actual threat actors. Instead, practitioners are left to deal with both threat actors and condescension from the labs.

llm议题 @0xTriboulet 原文 ↗

RT @ThePrimeagen: Imagine a world where instead of hiding every model behind closed doors for hacking we instead had things more opened. Perhaps we wouldn't have this week from hell.... Perhaps handing capabilities to a select few while attackers used abliterated models wasn't the best strategy... this almost feels intentional because now everything is going to be hacked and we are going to have literal weeks of chaos

llm议题 @0xTriboulet 原文 ↗

Based on public statements, it seems that open models have been the primary source of benefit to cybersecurity. So when you say “we tried to warn the world”, I don’t know what you’re talking about. The world tried to warn *you* that cybersecurity researchers in offense and defense need unfettered access to the best models possible. That is largely *still* not the case.

llm议题 @0xTriboulet 原文 ↗

There are an increasing number of YouTubers who I can no longer listen to because they use AI to write their scripts. I used to really enjoy their extemporaneous commentary, but now their videos are infected with the never ending verbosity and cringeworthy AI-isms

llm议题 @C2IRIS 原文 ↗

家用网关固件逆向的完整方法链,可复用到其他 IoT 设备。

Write-up on reverse engineering the Eufy HomeBase 2 firmware https://t.co/po4Fho7AfI #infosec https://t.co/nFKYqOHUE1

固件逆向iot报告 @0xor0ne 原文 ↗

解释 LLM 行为偏离预期的机制,对编写 agent 任务提示有直接参考。

RT @albinowax: Have you ever felt like a model isn't actually trying to do what you asked? Did you just ask it wrong, or is there something else at play here? In "The model isn't cooperating", I take a little look. Link below 👇 https://t.co/BKIMxLTaOn

llmai_agent @moyix 原文 ↗

钓鱼演练设计失当会推高凭据倒卖风险,安全意识项目需改口径。

RT @RachelTobac: I’ll tell you what I’ve told many customers before: Creating a phishing pretext like this is likely to backfire and create multiple insider threat issues. Folks may sell their credentials just to spite you for your cruelty. Of course criminals could use this pretext too, instead educate on it. Show an example of its likelihood rather than tricking people into thinking they’ll get a ring.

钓鱼内部威胁报告 @_xpn_ 原文 ↗

客户端通信层被滥用的实例,思路可迁移到其他桌面客户端。

> Be drankensang > wont be able to run on linux > if it runs on linux its via steam and its acc locked > be me > custom patch clients > exploit node bridge handler > exploit raknet > allow multi account login in game via steam rather than one account per steam acc > cheers. https://t.co/Q1tbStHFiI

漏洞客户端 @kernelstub 原文 ↗

去中心化网络充当隐蔽 C2 通道,给封堵与检测带来新盲区。

ESET researchers Anton Cherepanov & Peter Strýček uncover MythicRatatoskr – a previously unknown Go-based backdoor hiding its C&C server inside the Yggdrasil mesh network – which paved the way for a destructive attack against a Ukrainian grain firm. https://t.co/XnGLQj9Bd5 https://t.co/JKwIbuglSJ

事件报告供应链 @virusbtn 原文 ↗

伪装常用安装渠道,开发者工作机是直接命中目标。

The Zscaler Threat Hunting team identified an SEO poisoning campaign that distributed Armorsteal through websites impersonating Chocolatey and the Microsoft Activation Scripts (MAS) tool. https://t.co/RftHjAGF07 https://t.co/Fs3cp1bTcx

事件供应链报告 @virusbtn 原文 ↗

代码基与 StealC 相同,旧 IoC 与检测规则仍可复用。

RT @varonis: Four months after Operation Endgame, StealC's seller is back. Dubbed Stadia, this new product is a full RAT with hidden desktop control, a command shell, a file explorer, and a loader. Varonis Threat Labs got the build the day it was listed, reversed it, and ran its panel against a host of our own, discovering that the code inside matches StealC's. Operation Endgame forced StealC's seller to start from scratch with new servers, names, and panels, but it's clear they're not done yet. Read more about our findings here: https://t.co/0WOqmpMcLX

事件报告移动端 @virusbtn 原文 ↗

假验证步骤降低门槛,普通用户即可被诱导落地载荷。

RT @HuntressLabs: Hungry? This ClickFix campaign relies on some mouth-watering domain names: grillmasterspro, EmbersAndBites, and more. Read more about an attack that starts with a Google search for Amazon business and ends with a serving of “finger-licking” malware: https://t.co/aJ4ShqoXBL https://t.co/w3QSysYh4k

事件钓鱼报告 @virusbtn 原文 ↗

RT @MalwareTechBlog: I'll be the first to admit I'm mediocre at exploit development, but I am in contact with enough extremely talented exploit developers to say this for certain: Anyone claiming they were "easily" able to generate a working php2shell exploit with AI is either: a) a world-class exploit developer with an extreme case of imposter syndrome. b) currently looking at some very confident-sounding but wrong LLM output. I'm rather certain it's the latter based on the number of AI-hallucinated exploits I've seen posted on GitHub. Threat actors are also blasting my honeypot network with "php2shell" exploits that not only don't work, but aren't even within 10,000 miles of anything to do with how the exploit actually works. I've seen the same technique from the attached image in multiple alleged "php2shell exploits", including one on GitHub with over 100 stars. Problem is, it's not php2shell, nor is it even a vulnerability at all. It's a technique known as "session poisoning", which can be leveraged to exploit local file inclusion (LFI) vulnerabilities in PHP apps (not PHP itself). The only thing it has in common with CVE-2026-109618/CVE-2026-109619 is they both use the "PHP_SESSION_UPLOAD_PROGRESS" feature. And mitigation advice for both issues reference setting "session.upload_progress.enabled" to "Off". People are shoving loose details of the vulnerability into LLMs, the LLMs are keying in on the "PHP_SESSION_UPLOAD_PROGRESS" string, then generating session poisoning code that has nothing to do with the actual CVE. And because the people doing this have absolutely no idea what they're doing, they're confident enough to post their "proof of concept exploit" to GitHub and share it on social media. So if you see people talking about "session poisoning + LFI" it's likely their LLM hallucinated an exploit based on a 2018 blog by Orange Tsai talking about a local file inclusion attack vector. What's even worse is that "security researchers" are then using LLMs to triage these exploitation attempts, falling victim to the same hallucinations, then confirming there is ongoing mass-exploitation (there currently isn't). The actual real exploit is very difficult to exploit, and so far I've not seen anyone successfully get it working on more than a handful of PHP version with highly-specific configurations. LLMs are useful tools in the hands of skilled professionals, but for others, they're often just a Dunning-Kruger accelerator.

llmpoc @artem_i_baranov 原文 ↗

RT @S1r1u5_: if you are arguing that we should rapidly proliferate these capabilities through open models so defenders can find and patch vulnerabilities, while accepting that there will be a period of disruption, but also recognizing that committing crimes has real world constraints, including going to jail, that makes sense. there's a reasonable argument to be made there. or if you're arguing that we shouldnt rapidly diffuse these capabilities, and instead give defenders and vetted security researchers access to the most capable models first to get ahead of attackers, because there will be catastrophic events due to proliferation, that also makes sense. but pretending these models can't do shit and that nothing fundamentally changes is irresponsible. we can disagree on how it should be distributed, but i think it's far more productive to reason through how rapid diffusion and slow diffusion might play out, what the tradeoffs are, and update our views as evidence comes in, rather than turning this into another politically polarized debate discarding what models can do.

llm议题 @S1r1u5_ 原文 ↗

RT @SinSinology: next time you're thinking about researching the linux kernel (@linuxfoundation), remember that someone on the receiving end of your report might talk about you like this: in one of the threads, kernel maintainer "Peter Zijlstra" decides to call the ZDI team "monkeys" there's a longer argument about how ZDI submits reports, whether they should include patches, and how fixes get tested. fair enough, have that discussion but how does insulting the people reporting vulnerabilities help? i thought the kernel's code of conduct was something everyone on these lists was expected to follow sad to see people spoken to like this while trying to help improve linux security. crime scene: https://t.co/4Ud7JUz10M https://t.co/vXiHlvyZGi

内核议题 @FuzzySec 原文 ↗

𝗛𝗼𝘄 𝗯𝗶𝗴 𝗶𝘀 #php2shell 𝘀𝘂𝗿𝗳𝗮𝗰𝗲 𝗮𝘁𝘁𝗮𝗰𝗸 𝗮𝗿𝗲𝗮? 😅 ​ What 838 actually means? DeviceTvmSoftwareVulnerabilitiesKB is essentially vulnerability knowledge-base data. After mv-expand AffectedSoftware, the query counts the expanded affected-software entries associated with that CVE. ​ #VulnerabilityManagement

cve报告 @0x534c 原文 ↗

RT @ESETresearch: #ESETresearch uncovered #MythicRatatoskr, an undocumented Go backdoor hiding its C&C server inside the #Yggdrasil mesh network. It was used against a Ukrainian 🇺🇦 grain logistics firm in an intrusion that ended in a destructive #Sandworm attack. https://t.co/cdMywzNBtf 1/6

事件报告 @virusbtn 原文 ↗

RT @ImposeCost: Years ago I challenged our community to do a better job at characterizing exploitation "in the wild." There's lots of reasons for why, but the primary reason at the time is security researchers were using "in the wild" as the signal that it was low to no risk to publish offensive capability. "In the wild exploitation" at one victim in one country is not anywhere near the same "mass exploitation." PHP2SHELL is a good example of where there are private exploits that are not publicly available, and there are less potent exploits that are, and there's mass exploitation *attempts* using these less potent exploits and they aren't really working in real scenarios relative to the volume of attempts. Does that matter? It does. It's bought a lot of time for down stream would be victims to put durable mitigations in place.

议题poc @0xTriboulet 原文 ↗

关联式检测可以直接实现为自定义规则并触发自动隔离,附有共享规则链接。

🎯𝗔𝘁𝘁𝗮𝗰𝗸𝗲𝗿 𝗜𝗣 + 𝗩𝘂𝗹𝗻𝗲𝗿𝗮𝗯𝗹𝗲 𝗣𝗛𝗣 = 𝗛𝗶𝗴𝗵-𝗖𝗼𝗻𝗳𝗶𝗱𝗲𝗻𝗰𝗲 𝗗𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻 ​ 🌐 DeviceNetworkEvents → Remote IP matches a GreyNoise-tagged 𝗽𝗵𝗽𝟮𝘀𝗵𝗲𝗹𝗹 𝗮𝘁𝘁𝗮𝗰𝗸𝗲𝗿 🛡️ DeviceTvmSoftwareVulnerabilities → The same endpoint has a 𝘃𝘂𝗹𝗻𝗲𝗿𝗮𝗯𝗹𝗲 𝗣𝗛𝗣 𝗶𝗻𝘀𝘁𝗮𝗻𝗰𝗲 ​ Put them together: 𝗔𝘁𝘁𝗮𝗰𝗸𝗲𝗿 𝗶𝗻𝗳𝗿𝗮𝘀𝘁𝗿𝘂𝗰𝘁𝘂𝗿𝗲 → 𝗩𝘂𝗹𝗻𝗲𝗿𝗮𝗯𝗹𝗲 𝗲𝗻𝗱𝗽𝗼𝗶𝗻𝘁 → 𝗛𝗶𝗴𝗵-𝗰𝗼𝗻𝗳𝗶𝗱𝗲𝗻𝗰𝗲 𝗱𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻 🔥 ​ When both conditions are observed, the signal becomes significantly more actionable for defenders. This is a good candidate for a 𝗰𝘂𝘀𝘁𝗼𝗺 𝗱𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻 + 𝗮𝘂𝘁𝗼𝗺𝗮𝘁𝗲𝗱 𝗱𝗲𝘃𝗶𝗰𝗲 𝗶𝘀𝗼𝗹𝗮𝘁𝗶𝗼𝗻, helping defenders move quickly from threat intelligence to containment while investigation and remediation take place. ​ PHP2Shell ThreatLock KQL: https://t.co/wJvaqP9yhN ​ #Threathunting

检测规则php事件响应 @0x534c 原文 ↗

覆盖 shellcode、进程注入与内存扫描规避,是搭建实验环境的现成教材。

RT @cr3ghost: Proper old-school hacker energy. If you're into Windows internals, exploit development, malware memory or reverse engineering, Forrest Orr's (@_ForrestOrr) work is absolutely worth bookmarking. The FREE archive covers: - Windows memory corruption - Stack overflows + ROP - DEP, ASLR + CFG - Shellcode development - Windows internals - Malware memory artifacts - Phantom DLL Hollowing - Process injection + hollowing - Memory scanner evasion - Moneta memory analysis - PE internals - Browser exploitation - Chrome UAF exploitation - Firefox / IE exploit chains - Sandbox escapes - Defensive memory analysis And unlike a lot of modern security content, there are actual tools, PoCs, shellcode and exploit-development experiments to go with the research. Blog: https://t.co/cStzeIDi7t ExploitDev repo: https://t.co/JKaqBFO1QO Some of this is older material, but that's part of what makes it good. Proper fundamentals and low-level research that still teaches you how the machinery actually works. Huge credit to Forrest Orr (@_ForrestOrr) for keeping it publicly available. #ExploitDevelopment #WindowsInternals #ReverseEngineering

逆向内核工具 @cr3ghost 原文 ↗

从 NTLM relay 到 PRT 滥用的完整路径,可作为身份攻防的系统学习清单。

RT @cr3ghost: If you're into Active Directory, Entra ID or identity security, bookmark this. Researcher @_dirkjan has built one of the best FREE identity security research archives online. The rabbit hole covers: - Active Directory internals - Kerberos - NTLM relaying - Forest trusts + SID filtering - AD CS - mitm6 + krbrelayx - Zerologon - BloodHound - Entra ID internals - Primary Refresh Tokens - Windows Hello for Business - Conditional Access bypasses - OAuth + token abuse - Cloud Kerberos Trust - AD -> Entra lateral movement - Entra -> AD lateral movement - Temporary Access Pass abuse - Federated credentials - Managed Identities - Intune certificate abuse - ROADtools + ROADrecon + roadtx - Actor Tokens Some absolute classics: - Relaying Kerberos over DNS - NTLM relay to AD CS - Active Directory forest trust bypasses - Domain Admin from Entra via Cloud Kerberos Trust - Phishing Primary Refresh Tokens - Windows Hello key abuse - Conditional Access bypasses - Global Admin in every Entra tenant via Actor Tokens There are also years of FREE conference slides, demos and presentations. Blog: https://t.co/8qEBsLuiPy Talks + slides: https://t.co/5neZc1JAnQ Huge credit to @_dirkjan / Dirk-jan Mollema for publishing so much deep identity research openly. #ActiveDirectory #EntraID #RedTeam

active_directory云报告 @cr3ghost 原文 ↗

为“已被利用”的说法提供相反的判断依据,是判断补丁紧急程度的参考。

**disclaimer, delusions may still apply (obviously between all the different versions and environments the UAF memory corruption shit varies) (( still not yet seeing (or aware of) any actual real in-the-wild exploitation or post-exploitation beyond scanning or failed attempts)) https://t.co/RuCxEBLz02

phpcve事件 @_JohnHammond 原文 ↗

APT 已把 AI 诱饵和攻击安全软件写进攻击链,防御方需重估自身工具链。

RT @hypen1117: While recent vulnerabilities have been dominating the headlines, APT actors continue to operate quietly under the radar. And #Lazarus group has been busy. At the #TheSAS2026 conference, we shared our latest research on #OperationFlashHole, revealing a more sophisticated attack chain and an evolving malware arsenal to target South Korea 🇰🇷. 𝗦𝗼𝗺𝗲 𝗸𝗲𝘆 𝗳𝗶𝗻𝗱𝗶𝗻𝗴𝘀  ➡️ Targeting security software, AI-assisted lures, complex malware framwork, post-quantum cryptography. We also examined overlaps across Lazarus operations to better understand their recent activities. Enjoy the read! 🔗 https://t.co/OXy8YnJyNk

aptai_agent事件报告 @unpacker 原文 ↗

AI 找漏洞的能力边界之争,决定后续该往哪投入。

uhm... i think this seems like the wrong approach, especially if the goal is to formally verify every piece of software we have and eliminate every bug, we should definitely formally verify lots of software but "everything"??? first of all, how do you even deal with state explosion and the sheer complexity of real world software and its dependencies? are you really sure mythos 8 or gpt-8 won't find flaws in your proofs, specs, or assumptions? how do you even specify the behavior of nondeterministic agents, especially you have to balanced against usability? i don't think the goal should be eliminating vulnerabilities or proving their absence. that's not going to happen at the scale and complexity of real world software. start by assuming vulnerabilities are inevitable, then build around that assumption. reduce attack surfaces, isolate critical systems, and make exploitation as difficult as possible. i think thinking cyber is just fixing vulnerabilities might not be a great idea.

形式化验证llm议题 @S1r1u5_ 原文 ↗

把「AI 能否根除漏洞」变成可检验的公开预测。

RT @S1r1u5_: i think intuitively cyber is fundamentally offense dominant, but let me put my money where my mouth is. want to wager? by the end of 2028, i’ll concede if we formally verify most of the critical software we rely on like chromium, firefox, the linux kernel, nginx, next.js, and apache. not even asking about composition or applications built on top of them, just the core software, including every dependency. alternative wager, maybe for you and @JasonDClinton frontier labs have absurd amounts of tokens. by the end of 2028, if they formally verify every dependency that codex or claude code relies on(note agent actions should be part of the spec), including cloud infrastructure, desktop clients, and their servers, i’ll concede. actually, a much more arrogant wager, i would bet that even after 2028, me, my team, or some other security researcher will still find a software vulnerability that compromises a critical asset at vercel or anthropic and responsibly disclose it. i’d love to lose this bet.

形式化验证漏洞议题 @S1r1u5_ 原文 ↗

RT @ShangguanJiewen: China is the global humanoid robot industry and everyone else is just watching. Chinese companies accounted for 95%+ of global humanoid robot shipments in the first half of 2026. 19,000 humanoid robots, and scaling is rapidly accelerating. Try to imagine how will eventually sell at the scale of cars. This is a Ford motor moment. Chinese companies will DOMINATE personal humanoid robot manufacturing, defining the 21st century. Millions of homes will soon have humanoid robot helpers. Then tens of millions. Then, like cars, or washing machines, these will be EVERYWHERE. And China is at the forefront.

机器人供应链 @0xocdsec 原文 ↗

可本地部署的 Agent 模型,攻击性安全基准成绩高,便于自建评测

RT @MiniMax_AI: MiniMax M3.1 Flash weights are now public. 428B parameters total, 23B activated per token. Agentic performance: A generational leap in agentic tasks, coding, and cybersecurity — 76.2 on BankerToolBench, 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, 29.6 on Agents' Last Exam, and 82.2 on CyberGym. Efficient by design: MiniMax Sparse Attention and DSpark speculative decoding improve inference efficiency. Hugging Face: https://t.co/JDYA1zZh54 ModelScope: https://t.co/DQcmOjTwhO API: https://t.co/fHRdSV7BwZ M Plan: https://t.co/LYnYZcsesu MCode for desktop: https://t.co/SGmLTI3fFU

llmai_agent工具 @0xocdsec 原文 ↗

本地调试 Agent 出网流量与 TLS 明文检查的常用工具

mitmproxy v12.3.0 — An interactive TLS-capable intercepting HTTP proxy for penetration testers and software developers. https://t.co/2cA2t9rsN8 https://t.co/DiUbI3Xem6

工具代理tls @KitPloit 原文 ↗

解释漏洞可利用性与经济激励,有助于判断哪些洞会被规模利用

This is 100% correct when looking at it from a single-system perspective. ASLR does not create a significant barrier to successful exploitation. But when looking at it from a mass-exploitation perspective, there are a lot of hidden complexities that come into play. Vulnerability research is one of the few fields where you tend to make more money on the right side of the law. There isn't a huge demand for black market exploits. Most of the money is in bug bounty, working as a vulnerability researcher, or selling to grey-market exploit brokers. This means that even the smallest bit of added complexity can take a vulnerability from 'cybercriminal free-for-all' to 'unlikely to be widely exploited'. Mass-exploitation events are almost always preceded by someone publishing fully-weaponized point-and-click exploit code. It's extremely rare for criminal groups to successfully develop exploit for higher-complexity vulnerabilities on their own, even with AI. So, less researchers capable of exploiting the bug means less chance of a working weaponized exploit being posted online. But also, the increased difficulty of exploitation, thus increased time/effort, discourages those who can exploit the bug from giving it away for free. There's also per-target variability (which I don't know for sure is the case with this bug). But with more complex bugs, it'll sometimes be impossible to make universal exploits. In extreme cases, attackers may need to build exploit for each specific system or software version, or each specific config, or maybe even both. So, while it's correct that it's not hard to find an ASLR bypass, it is difficult to find a bypass that works reliably at scale. So far, most of the working exploits I've seen are limited to just a handful of PHP versions, or bruteforcing ASLR. And that makes all the difference when it comes to the risk of threat actors spraying exploits at the entire internet. For this specific exploit, the honest answer is, I don't know. It's above my pay grade. But I'm almost 100% certain that the level of bad this will be will be directly dependent on what kind of exploits security researchers make public.

议题利用aslr @MalwareTechBlog 原文 ↗

RT @0xp1ain: Researcher A discovers a critical vulnerability using IDA and pwntools → This is recognized as Researcher A’s technical achievement, made possible by their ability to effectively use the right tools 🎉 Researcher A discovers a critical vulnerability using an LLM → This is also Researcher A’s technical achievement, made possible by their ability to effectively apply the right heuristics and tools. So why is it that, in this case, people say the discovery was only possible because of the LLM or the harness? 🤣 Simply using an LLM doesn’t change who deserves credit for the technical achievement. of course, LLMs can enable researchers to achieve performance or speed improvements that were previously impossible, sometimes by 5x or even 10x But when an extremely difficult, critical vulnerability is discovered, I believe the credit should go not to the LLM itself, but to the human researcher who skillfully directed it

llm议题 @0xocdsec 原文 ↗

RT @HackingLZ: We’re going to look back at this era when LLMs have supposedly solved software security, only for people to realize how much money is still being lost every year to BEC and other attacks that require little technical sophistication. All while LLMs have arguably become the best BEC tooling on the planet.

llm钓鱼 @0xTriboulet 原文 ↗

RT @HackingLZ: Something that gets lost in all these conversations is that there were years when a major 0day would drop and technical details were almost impossible to come by. Multiple people here would stay up for days reverse engineering the issue and stitching together a working PoC, not just to prove exploitation was possible, but to help defenders understand the actual risk, develop IoCs, build detections, and clean up compromised environments.

议题poc @nas_bench 原文 ↗