推特安全流

可能涉及重要漏洞或技术分析，需点击查看。

RT @bitfalls: https://t.co/IUOPO4dkoD

恶意软件分析技术，对检测有参考价值。

RT @hasherezade: My research blog about #JSCeal is finally out! It was quite a journey. I hope you will like it! // #V8 #bytecode #malware #deobfuscation

AI 账户安全事件，影响使用 AI 服务的用户。

RT @TheRegister: Anthropic cracks down on hijacked user accounts mining AI tokens https://t.co/Uum3fJrE5l

影响安卓设备底层安全，研究价值高。

Exploiting Qualcomm BootROM to bypass the secure boot chain (slide deck, BH Asia '26) https://t.co/X0cBoWQLIA #infosec https://t.co/7V3nU7ZgVQ

可能涉及新漏洞，值得关注。

Hi, will be releasing some 0days for omarchy :).

涉及主流安全软件，PoC 公开需关注。

RT @cr3ghost: Two more fresh 0-day drops from @MSNightmare2000. GreenSection: NVIDIA user-mode GPU components, memory corruption / OOB write. PrettyPrague: Avast / Gen Digital sandbox EoP to SAM dump + SYSTEM. Red teamers: BYOVD isn't the only path to privileged execution. Blue teams + threat intel: public PoCs shorten the path to operational abuse. Start understanding the telemetry before these techniques show up in tooling. https://t.co/qIAoCBYUtY https://t.co/oqHEV7uEnf #0day #VulnerabilityResearch #ThreatIntel

内核安全研究资料，对逆向有帮助。

I’ve started my PatchGuard research. Here are the first notes I’ve taken. I’ll keep updating this document over the next few days https://t.co/1XCaFjcF83 The main goal of this research is to document all the structures monitored by this protection, at least for my Windows build

供应链攻击真实案例，警示开源依赖风险。

RT @USEmbAustralia: A joint @FBI, @AusFedPolice and @WA_Police investigation has led to the arrest of two Western Australian men charged with allegedly hiding malicious code in open-source software, harvesting 500,000+ credentials and compromising 1,000+ organizations worldwide. A powerful example of U.S.-Australia law enforcement cooperation and our shared commitment to combating the growing threat of cybercrime. https://t.co/sT0LC7r9na

现代 bootkit 技术，绕过 DSE，研究价值高。

RT @5mukx: RedLotus is a Rust based Windows UEFI bootkit that executes before ntoskrnl.exe to bypass DSE and manually map unsigned kernel drivers. A research focused PoC demonstrating modern bootkit techniques and low level rust development. by MemN0ps Blog:- https://t.co/YKY5KN6DAK https://t.co/w1HsZHF3GD

新内核架构，可能影响未来安全研究。

RT @Itsfoss: Multikernel has got its first public release, but it isn't geared towards desktop users. https://t.co/9sOUzOqeow

学习加密实现的好材料。

RT @oneandonlyhusam: ultralight-AES The smallest implementation of AESin C I can think of. Made as a follow up to my tweet yesterday. This implementation is of AES-128 in ECB mode so its not really the most secure implementation, that would be CBC. Goal was simplicity. https://t.co/4abzsGEMCM

AI 辅助漏洞分析案例，有参考价值。

RT @Kevin2600: https://t.co/2J8asu4Fzs

真实攻击案例，展示 MFA 绕过手法。

The user kept getting MFA prompts he did not request. He denied all of them. Normally, you would assume somebody had his password and was just trying their luck. But this one felt different. We checked the sign-in logs and saw several attempts against his Microsoft 365 account. His password was changed, sessions were revoked, and we moved on. Then the MFA prompts started again. That was when we decided to look at the laptop properly. While checking the running processes, I found: WebViewHost.exe The name looked believable because Windows and Microsoft applications use WebView components all the time. But this one was running from: C:\Users\michael\AppData\Local\Microsoft\WebView\WebViewHost.exe I checked the signature. Unsigned. That already looked bad. Then I checked persistence and found a scheduled task called: Microsoft WebView Update It was launching the same executable. So now we had an unsigned Microsoft-looking binary that was starting automatically. I went through the Sysmon logs next. Around the same time the program appeared on the laptop, it created temporary files called: login_db.tmp and cookies_db.tmp That immediately made me think about browser data. I checked the process network activity and found it maintaining an HTTPS connection to an external IP address. At that point, we isolated the laptop. We could not say from two filenames alone that every password or cookie had definitely been stolen, so we preserved the files and the executable for proper analysis. But the combination was enough to treat it seriously. Unsigned executable. Fake Microsoft name. Persistence. Browser-related temporary files. Outbound connection. When we spoke to the user, he remembered installing what he thought was a PDF tool the previous evening. He needed to sign a document quickly, searched online, downloaded one of the first tools he found and installed it. The PDF tool worked. That was why he never suspected anything. Later analysis confirmed that another component installed alongside it was collecting browser information. That explained why simply changing the password was not enough for us. If an attacker has stolen active browser session data, the investigation has to go beyond just resetting credentials. You need to revoke the sessions as well and deal with the endpoint that leaked them in the first place. The interesting part was how normal the malware tried to look. WebViewHost.exe Microsoft WebView Update Even the folder path had Microsoft inside it. If you only look at names, it can fool you. Always check the path, signature, persistence and behaviour. That is usually where the real story is.

域名侦察工具更新，提升效率。

Findomain v11.0.0-beta.1 — The fastest and complete solution for domain recognition... https://t.co/BD4JKBzf4u https://t.co/kYLZ52cU8D

AI 安全系统竞赛结果，展示技术趋势。

RT @trendai_RSRCH: In the CyberGym competition for AI security systems, engineering and persistent knowledge layers drove success. The TrendAI™ exploit-remediation engine ranked #1 by leveraging the ZDI catalog’s historical intelligence for every exploit attempt. Read more: https://t.co/2oujnj4uyM

政府机构遭勒索，影响公共服务。

RT @DailyDarkWeb: 🇩🇪 BERLIN GOVERNMENT HIT BY CYBERATTACK — ATTACKERS NOW DEMANDING RANSOM The State of Berlin has confirmed that attackers behind the recent cyber incident affecting Berlin's government network are attempting to extort the city. Berlin's Governing Mayor Kai Wegner and Interior Senator Iris Spranger responded: "The State of Berlin will not be blackmailed." The latest forensic investigation has also uncovered additional data exfiltration from systems within Berlin's Senate Department responsible for mobility, transport, climate protection and environment. * Data exfiltration occurred between August 7–12, 2026 * The affected government networks were disconnected on August 14 * Investigators are still determining the content and volume of stolen data * Authorities cannot rule out exposure of personal information * Other non-public government information may also have been compromised * Berlin's State Criminal Police Office and public prosecutors are investigating * German federal security authorities are participating in the investigation * Germany's BSI and Berlin's data protection authority are being continuously informed * Forensic examinations and scanning of Berlin's government network remain underway * Berlin's ICT emergency response team remains activated ⚠️ Analyst Note: This incident has now moved beyond a conventional government network compromise. Berlin has confirmed both data exfiltration and an attempted extortion operation. The fact that attackers maintained the ability to exfiltrate information over several days before affected networks were disconnected makes the continuing forensic investigation particularly important. The government has not yet publicly identified the threat actor, disclosed the ransom demand, confirmed ransomware deployment, or revealed exactly what information was stolen. Until attribution is established, any ransomware-group or dark-web claims connected to this incident should be treated as unverified. Source: State of Berlin — August 28, 2026 https://t.co/GGKB2BsqEA #DDW #CyberSecurity #Berlin #Germany #DataBreach

可能影响 macOS 用户，需关注后续。

RT @mufeedvh: this is a very exciting macOS vuln but how long do i wait man. it's been like a month now. https://t.co/VXHf2t3Kmz

EDR 内省工具，可能有助于检测绕过研究。

RT @0xbfho: boh starred evilele/EDR-Introspection on Github https://t.co/PPv4Wn8eIG

Windows 内存分析教学，适合学习。

RT @zodiacon: New video: Process Memory Map in Code (Part 3). MemMap now enumerates threads, finds each TEB via NtQueryInformationThread and PHNT, and reads thread stacks with ReadProcessMemory. Full write-up: https://t.co/RbuapZuYAn

恶意软件新功能，展示地理定位攻击。

my rootkit/c2 list all wifi from the hacked computer and send it to the c2 and it map out the location where the wifi is located and map them out and try to find live cams close to the target. all automated ... STILL UNDER DEV .. it will be apart of my c2/rootkit when its done. https://t.co/sAyIHhgRk9

云原生安全网络组件更新。

cilium v1.20.1 — eBPF-based Networking, Security, and Observability https://t.co/f7FVXiu3zL https://t.co/TdxhyoaAch

AI 辅助逆向工程实战数据，有参考价值。

RT @cr3ghost: 235 BILLION tokens later, @momo5502 went back and audited what the AI agents actually did while decompiling Modern Warfare 2. ~2 GB of session logs. Where the tokens went. Where the agents wasted time. What actually worked. What failed. What he'd change next time. This might be even more interesting than Part 1. https://t.co/U1a3srbjqm #ReverseEngineering #AgenticAI #GameHacking

移动安全改进，了解即可。

RT @The_Cyber_News: Android 17 Adds New Protection for Wi-Fi Tracking to Keep Your Device Secure More Details: https://t.co/ShbvxzBQ6A Google has introduced new privacy and network-security protections in Android 17, including stricter controls that prevent apps from scanning devices connected to a user’s local Wi‑Fi network without permission. The update is designed to reduce household profiling, limit tracking, and defend users against threats ranging from malicious Wi‑Fi activity to cellular SMS scams. Android smartphones often connect to home Wi‑Fi networks shared by smart TVs, cameras, game consoles, speakers, printers, and other Internet of Things devices. Previously, an app could potentially discover devices on the same local network without clearly asking the user for access. #cybersecuritynews

对抗 AI 视觉识别，趣味性大于实用性。

RT @VaibhavSisinty: A Berlin artist made a shirt that makes you invisible to AI surveillance cameras. Like, the camera sees everyone else but skips right over you. The pattern looks like a loud Hawaiian shirt to humans. But to AI detection systems, it breaks the outline of a human body so completely that the model can't register a person is there. He built it by testing patterns against YOLO until the AI stopped seeing a human. Now he's selling it for $75. If his name sounds familiar same guy who tricked Google Maps into showing a fake traffic jam by pulling 99 phones through an empty street in 2020. It won't fool every system. But the government systems watching you are closed, proprietary, and impossible to test independently. The shirt is the question. The system that can't be questioned is the problem.

AI 训练动态，间接相关。

RT @rohanpaul_ai: POV: somewhere inside OpenAI. truckload of Macs are training computer-use-agents. https://t.co/6Dph2WogJ1

K8s 策略引擎更新，常规维护。

gatekeeper v3.23.1 — Kubernetes policy engine with OPA-based admission control, mutation, and audit for enforcing security and compliance... https://t.co/kPiOsrma7n https://t.co/BrUFTxHG9I

影响广告拦截器，但非安全核心。

RT @chromeunboxed: Google has officially deleted all Manifest V2 extensions from the Chrome Web Store today. Here is what happens to your extensions and ad blockers. https://t.co/fINk5s31Ru

内核堆利用学习资源。

RT @MrT4ntr4: Hey, I just published a book for learning Windows Kernel Segment Heap Internals and exploitation techniques based on my own notes. Read it here: https://t.co/Nf1Y0qzHVm There's a lot to wrap your head around, so I started creating my own diagrams and explanations to make it easier to understand. Hope it helps! :)

代码注入技术，红队可参考。

Module Overloading - Select the DLL - Open the DLL - Create an Image Section - Map the Section - Change Memory Protection - Overwrite the Image - Restore Memory Protection - Execute the Payload https://t.co/uxJlZO6JNY

AI 安全社区动态，了解即可。

People say that the OpenAI swarm agents can't be compared to humans, but let's look at the facts on the ground: they put a bunch of infosec nerds together, gave them impossible KPIs, and a within a few months they generated huge amounts of drama

与前述研究重复，略读即可。

RT @XorNinja: Just dropped new research and some dope artwork from Lukas Maar and @lcamtuf: OEMpocalypse Now! Check it out: https://t.co/qQSjJdVH5L While you’re there, take a look at our new research homepage, designed by yours truly: https://t.co/XurBelzflI OK, I lied. Claude designed it. But I pressed “yes” many times.

娱乐内容，非核心安全。

😎 Game Hacking Shenanigans ☑️ Short and Simple Explanations ✅ Epic Music https://t.co/PklCxCwDpH

具体漏洞公告，影响面有限。

ZDI-26-607 is out. Smooth process with ZDI/MSRC. The HTML injection was the entry point. The trust graph was the interesting part. Advisory: https://t.co/H9DQkn58pg https://t.co/OyDjTl5vog

行业倡议，非具体技术。

RT @OpenAI: We have a limited window to strengthen cyber defenses, and together with organizations including @AnthropicAI, @awscloud, @Google, @Microsoft, and @Oracle, we're calling for a global effort to give defenders the tools, resources, and support to protect the infrastructure we all depend on. If we act decisively, we can turn today's AI advances into lasting improvements in security and make our digital world safer for everyone. https://t.co/f33JRVCiJb

无信息量。

bro.. 🎈 https://t.co/6b9KPRKE0U

与前述 PaperCut 相关，略读。

RT @watchtowrcyber: Death by a thousand (paper)cuts (also known as WT-2026-0144) brings us back aboard the HellScape Express. The saga continues… and we'll be back soon. (We gently, kindly, and calmly suggest pulling PaperCut entirely off the Internet at this point) https://t.co/xa1o4XedrC

常规安全建议。

RT @PyroTek3: There are numerous ways to extract cloud credentials (like tokens) once an attacker gains access to a workstation. Protect your cloud admin credentials as Tier 0. https://t.co/yjD8tKv2eA

非安全内容。

RT @zodiacon: Open for work! My stint at Roblox has ended. Looking for interesting roles... if you know someone that is looking for someone with my skills, feel free to point them in my direction :)

会议信息，非技术内容。

I'm thrilled to give a talk in SEC-T this year! Would love to see you there :)

无关内容。

RT @NarvajaSyl46261: "CUATRO VOCES TRÁGICAS" NUESTRA NUEVA PRESENTACIÓN! Estaremos el 4 de septiembre a las 17 horas junto a Cristina Quarella en la Biblioteca Sarmiento de Quilmes, con ENTRADA LIBRE Y GRATUITA.Hablaremos de Alfonsina Storni, Delmira Agustini, Alejandra Pizarnik y Marta Lynch. https://t.co/cnLFHELaAm

AI 渗透测试相关，但非紧急。

THIS WEDNESDAY: @vincent_olesen & I host a webinar on: 1) Why auth is so wickedly hard in autonomous pentesting; 2) How real-world sites do auth in ways that would make Sir Tim Berners-Lee weep; and 3) How we nonetheless got pretty good at it! Join us! https://t.co/atfjPWvDC0

AI 漏洞挖掘趋势，但无具体细节。

I’m not a low-level reverse engineer. I never mastered x86 or spent nights writing shellcode from scratch. Last week, I used AI + standard tooling and chained together a critical 0-day in a massively popular library. Years of memory corruption expertise, wiped out by a good prompt and the right skills. Respect to the OGs, but the game has fundamentally changed.

游戏安全相关，非核心。

👨‍🔬 Building An Anti-Cheat For Open Source Games • PEB anti-debugging • CRC32 integrity checks • Thread creation hooks • DBI framework detection 👉 https://t.co/D74Qgfm4oC https://t.co/Mp2u2hXHwX

硬件调试工具，小众。

hydrafw — Open-source firmware for HydraBus, a multi-tool for embedded hardware debugging, hacking, and penetration testing, supporting protocols like... https://t.co/YmmY3naCQW https://t.co/IoKssznu0i

通用安全建议。

A good reminder why always ry additional security layers like authentication app or passkeys to your social media accounts...when things move, they move fast and can be hijacked for not just spam, but intel gathering.

无实质内容。

Woot Woot, finally making stuff for my usage https://t.co/vXCVkyWLyz

车联网安全案例，但非核心。

The car kept waking itself up after it was parked The complaint sounded strange at first. The owner said the battery was fine, the alternator tested fine, and there was no obvious electrical fault. But every now and then, the car would sit for a few hours and still lose more charge than expected. So I started looking at what was happening after shutdown. The ignition went off. The doors locked. The cabin went quiet. But the vehicle network did not stay quiet. A few minutes after most of the CAN traffic dropped off, the bus became active again. Then it settled. Then it woke up again. That pattern repeated. At that point, I stopped thinking about the battery and started looking at which module was causing the wake-up. The gateway logs pointed back toward the infotainment side. That is where things got interesting. Modern infotainment systems are not isolated radios anymore. They can have Bluetooth, Wi-Fi, USB, cellular connectivity, navigation, app integration, microphones, cameras, and links into other vehicle networks through gateway modules. In this case, the infotainment unit appeared to be waking periodically and triggering additional CAN activity behind it. So I checked the network state, timestamps, and the identifiers that appeared immediately after each wake event. The same small group kept showing up. That gave me something useful to correlate. It did not mean the infotainment system had been compromised. A software bug, stuck Bluetooth process, failed update, misbehaving accessory, telematics event, or power-management issue could all create similar symptoms. But the important part was that the battery drain was no longer just “something electrical.” There was a repeatable network event behind it. This is one of the reasons automotive cybersecurity auditing can overlap so heavily with diagnostics. A good investigation is not always about finding an attacker. Sometimes it is about understanding why one computer inside the car is keeping several others awake when they should be sleeping. And once you can see that behavior in the logs and on the CAN bus, the problem becomes much easier to narrow down.

供应链风险案例，但无新细节。

Good break down of the built in 'ENDLESSDOORS' backdoors in the Chinese made home routers.

ATM 安全研究，但非核心。

RT @emptynebuli: I released 9 CVEs and tooling against the ATM supply chain at both @BlackHatEvents and @defcon this summer. I am pleased to share this was published in the following @WIRED article!! https://t.co/EJ6atOQbR5

招聘信息，非技术。

RT @andremoulu: 🚀 We're growing the @CellebriteLabs French team and hiring security researchers. RE/VR/exploitation/dev/fuzzing on Android/embedded systems and plenty of hard targets along the way. 🇫🇷 Looking for French-speaking candidates in France or EU. DMs open. RTs appreciated 🙏 Link⬇️

AI 模型动态，非安全。

RT @MiaAI_lab: you forgot to release Qwen3.8-35B. it's still not too late.

非安全内容。

RT @RuxandraTeslo: Unfortunately, 2 days ago, an mRNA vaccine trial in colorectal cancer, from BioNtech, has been interrupted due to lack of efficacy. Last week, Moderna stock went up ~ 160% on news that its mRNA cancer vaccine had positive results in a Phase 3 trial for reducing melanoma recurrence. I said at the time that we didn't cure curing cancer & that other cancers will be harder. I think this is a good occasion to discuss just how hard medical innovation actually is and why over broad claims of "curing cancer" are non-sensical most of the time. First, each cancer is kind of different. This is why one can't simply translate results from one type to another. We have plenty of examples: one is CAR-T therapies, which harvest a patient's immune cells and re-engineer them to attack cancer. They are curative in advanced forms of blood cancer, which is nothing short of miraculous. But they are much less effective in cancers that affect specific solid organs (eg pancreas). This mRNA cancer vaccine case provides another example: melanoma is a perfect target for cancer vaccines, because it is a so-called "immunologically hot" type of cancer. That means it's very easily reicognoscible and attack-able by immune cells, not least because it has many mutations, which create "neoantigens" that the immune cells can recognize. Given that mRNA vaccines work by further activating the immune system to react to these neoantigens, it's to be expected that they'll be more efficacious in immunologically hot tumours. Colorectal cancer, unless of a specific subtype, MSI positive, is not as amenable to immunological attack. So as a baseline, we should have expected a vaccine to be less effective. Now, there are other choices BioNTech made that probably made the difference. Most notably, that BioNTech tried the vaccine as monotherapy, whereas Moderna tried it as an adjuvant. What does that mean? The Moderna and BioNTech trials looked superficially similar: both used personalized mRNA cancer vaccines after surgery, when the visible tumor had been removed, to try to prevent recurrence. But Moderna’s KEYNOTE-942 tested its vaccine in melanoma together with pembrolizumab, an immunotherapy that removes "brakes" on the immune system, whereas BioNTech’s BNT122 colorectal trial tested the vaccine alone against watchful waiting. Moderna basically tested a more narrow and less impressive capability: whether vaccination could improve an already effective immunotherapy. Basically, the vaccine in this case generates tumor-specific T cells, while pembrolizumab removes an inhibitory brake on them. This is a nuance that often gets lost on people discussing cancer vaccines, who like to claim that they work alone! But BioNTech actually tested whether vaccination alone could eliminate residual colorectal cancer here, and it didn't work. Honestly, I'm not sure what BioNTech was thinking: testing both as immunotherapy and in a harder cancer. But this just goes to show how hard biomedicine is and how each victory has to be fought with teeth and nails!

硬件细节，与 MTE 相关但非核心。

RT @GrapheneOS: @roddux @viperbjk Pixel 6 used Exynos for cellular, Broadcom for Wi-Fi/Bluetooth and Broadcom for GNSS. Pixel 7a used Qualcomm for Wi-Fi/cellular but then they went back to Broadcom. Pixel 8a moved to Samsung for GNSS. Pixel 11 moved to MediaTek for cellular and GNSS. It's not a MediaTek SoC.

娱乐内容。

RT @TheCassiniGames: Graphics comparison - GTA 6 (base game) vs. GTA 5 (DLSS 5 + NVE) Technology is crazy! https://t.co/QZJJU8f01c

与前述 MTE 相关，略读。

RT @PavolRusnak: I remember buying Pixel 8 exactly because Google introduced ARM Memory Tagging Extension (MTE) in their Google Tensor G3 chip. Now Google really dropped the ball with Pixel 11. They removed MTE from Google Tensor G6 chip. Do not buy Pixel 11 and stick to Pixel 8, 9 and 10, until they reintroduce MTE back again. If you already bought one, return it and state "MTE missing" as the reason. Don't forget to run @GrapheneOS on your Pixel 8/9/10, because it is the only Android OS that uses MTE properly.

AD 攻击技术，但非核心。

RT @Synacktiv: In our latest article, @croco_byte presents new scripts allowing to simulate legitimate AD services in a flexible manner, and demonstrates their use through GPO exploitation ⬇️ https://t.co/qkgQsFltEa

AI 工具讨论，非安全。

This is very usable on real-world agentic coding! https://t.co/YY4s1tXiz9

与前述钓鱼活动相关，略读。

Mirror IOC (For those not on https://t.co/xwwu5qWnCR) https://t.co/wzp44tZ9ea

非安全内容。

Looking for a job.

项目总结，非新内容。

RT @hackyboiz2: [Wipeload Project ⛰️ — Step 9] Beyond the Chain (END.) Hi, this is the Wipeload team. Our Wipeload project has finally come to an end. To wrap things up, we’re sharing some final thoughts after this long journey. Thank you to everyone who walked this path with us along the way! We’ll keep working to bring you more interesting research in the future. :) Wipeload Project END. → https://t.co/AhOx8IRD5Z

教育资源，非安全。

RT @XenoKovah: Do you think everyone should have access to free professional-grade educational material regardless of race, gender, or geographic location? And that the material should be both open access and open source? That's what #OST2 is making happen!

AI 安全工具，但非核心。

aegis-latent-core — AI governance and evidence gateway for multi-provider LLM applications... https://t.co/eKCKQf8BR4 https://t.co/HHEp2i6gTX

无实质内容。

This is so ironic... https://t.co/KrECJny3V3

无实质内容。

@C5pider @bintracerlabs thanks man!

无实质内容。

@kyleavery @bintracerlabs great work kyle! this is honestly pretty cool :D

无实质内容。

guy who recommends active directory cause 'its harder to phish'

无实质内容。

@HackingLZ @techspence anyway, good bait. won't happen again

安全架构讨论，但非核心。

@HackingLZ @techspence you are way more likely to contain the blast on a workstation in Azure vs AD. so you moved the goalposts to just focus on initial access. but the whole advantage of Azure happens post initial access.

恶意软件开发，但无新细节。

I start working on sleeper cell style fucntion to my c2/rootkit so it sleeps until a specific date then it connects to the c2

编程教学，非安全。

Coding Huffman trees in C https://t.co/c8iOu6qvAA JB

AI 安全平台，但非核心。

agentZ — Zero-trust agentic AI platform. Supports SaaS and OnPrem (airgapped) deployments. https://t.co/JaG7yGXOWt https://t.co/NQzwC5yKEd

培训信息，非技术。

RT @pedrib1337: The baseband (a cellular modem) is one of the most complex and interesting embedded devices out there. Want to learn how to understand, debug, reverse engineer and pwn them? Come to my class @hexacon_fr, one of the best offensive cons in the world!

观点分享，非技术。

Some people in the community seem to be fully locked into the idea that there is no risk, no uplift, just hype from AI. I find this totally perplexing, it’s obvious how impactful the technology is. As capability progress continues rapidly we, more than other groups, should be advocating on behalf of people that don’t have the same foundational understanding.

AI 安全应用，但无新细节。

My 2 cents on AI adoption for threat intelligence and defense: I think the biggest value comes from integrating AI into our daily workflows rather than simply using it as a standalone tool. For example, the recently popular Rust crate ArrayRef was compromised(attributed to DPRK actor by some vendors) and used to deliver malware. The 2nd stage malware used DGA and selected its C2 based on the date. When I checked the domains for the relevant timeframe, they were registered through Namecheap and pointed to parking servers. I wanted to monitor these DGA domains on a daily basis, so I built a simple "loop" with Claude Code to automate the DNS registration checks. There are certainly other ways to monitor DGA domains, but AI made it much easier to turn a repetitive task into a simple daily workflow. If AI can take one repetitive task off my plate, I'll happily let it do the boring part while I focus on the interesting part.

非安全内容。

I Bought a ThinkPad Gen 2. Came with Windows. Used it for a month. Slow. Laggy. Garbage. Threw it in my room and forgot about it. A month ago, I installed Ubuntu on it. Holy shit. This thing is a monster. Windows was the bottleneck. Not the hardware.

观点，非技术。

RT @UK_Daniel_Card: 'great blogpost!'..... but actually it's not..... the #AI #madness is a sickness of the human mind! Brain worms for the digital age! A billion people who don't understand how computers work, watching Hollywood films, mad scientists who have forgotten how to science and a bunch of VC money making people who should know better, lie for cash! The entity..... is something from a film! The madness, is something from reality!

无信息量。

xD https://t.co/ggmVU2RCD6

非安全内容。

@0xpwnie @YogSoth0 @usetraceix Exactly bro donated 1k and now wants it back </3

非安全内容。

@YogSoth0 @usetraceix Wait what?? I thought they gave @kernelstub the funding for the purpose of reversing the flock cameras Why do they want to collect it back? Is there something i don't understand

非安全内容。

RT @UnoMasReactor: Including Chernobyl and Fukushima, nuclear is safer than wind and cleaner than everything For many of us, Josh included, nuclear is also responsible for both our nice-to-haves and our life support systems The world needs more nuclear energy https://t.co/Y32TwcZOOV

观点，非技术。

RT @robinebers: my favorite AI slop must be coding agents adding backward compatibility and legacy fallbacks for an app that isn't even deployed yet

观点，非技术。

RT @CraigHRowland: A big topic at Black Hat this year was AI safety and guardrails. There is no way it ever has a chance of working when Chinese and open weight models exist and don't care. I suggest cybersecurity teams arm up with the best and most unrestricted LLMs they can find and work around these AI vendors trying to put shackles on you.

无实质内容。

A bottom 5 take of all time https://t.co/ncdh5zzMOK

无实质内容。

Me: existing, thriving, vibing First Slot machine I sit next to: https://t.co/wLzGuE1rRq

纠纷，非安全。

@usetraceix trying to extort back money they sent on their will without Kernel asking for anything. These people are dangerous and toxic. Expose them any time you can. #scammers #traceix #criminals #extortion #x #support #cybersecurity #security #infosec #flock #exploit

无实质内容。

today I'm tweeting a lot because after a long time, I decided to take an break, do some random things, eat fruit-jelly-ice-cream and a long ride somewhere. i want to throw my phone somewhere else, at least for an day lol ! https://t.co/phjx35W9YF

非安全内容。

I told him to give me some time and he'll get it back, but as it seems that won't work out for him at all also lmfao. Guy dmd me every 2 days about it even though I told him give me time to take care of heredity stuff since my grandpa died. It's almost done kekw.

观点，非技术。

Everyone can use an llm. everyone can build things. But do you know the outcome of what you're building. 95% of people dont. thats what makes the difference when a technical and a non-technical person uses an llm. professionals know exactly what they want to build, why they're building it, and what outcome they want to achieve. using an llm is easy. knowing exactly what to build (you have the whole flow running inside your thoughts) , how to build it (your way of design), and why it matters is what makes a lot of difference. for example before 4 months i don't know too deep about compilers & stuffs. i started to refer llvm books, understand how things happens and done tried something in it ( and this is the coolest book ever ). i read books, a lot of books to understand stuffs, learn 1 or 2 different things to make things different. ( everyone cant post everything Problème de sécurité opérationnelle ) small story:- i have a friend of mine who works in xyz red team lead. his junior team mate got same model, same harness. Junior found nothing, while my friend found five 0 days in an month, reported himself, got a large sum amount and started a small startup.

观点，非技术。

I think there is a lot of truth to this. In cyber security but also surveillance awareness. Many scenarios, at least in the past, where accepted as too 'theoretical' as the amount of data that needed to be processed, entities cross referenced and analyzed was just too large for an ordinary threat actor to do something meaningful with it. But now with the support of AI, ever growing processing power, plenty of open-source information and more available storage space those "fringe" scenarios of the past are increasingly moving into the reach of smaller players too.

硬件安全细节，但非核心。

RT @lyq_sqsp: Hardware-based mitigation needs more time to deploy. For example, the @Starlink user terminal is the first non- smartphone IoT device that I've actually seen to have ARM PAC enabled (only in the kernel tho).

非安全内容。

I don't understand. Can someone please explain ? Is the Falcon license difficult to get ?

非安全内容。

Why do I always feel guilty when closing down an ai chatbot session without saying good-bye? Same for you? JB

无实质内容。

This is amazing! Well done! 😂

无信息量。

RT @WholesomeMeme: https://t.co/ygIXfK6VdC

娱乐内容。

RT @martn3ws: GTA6 vai precisar de muito pra superar o San Andreas, o mais icônico de todos kkkk https://t.co/0gp59McYMn

无信息量。

RT @syndrowm: "kill all humans?" "kill all humans!"

无实质内容。

I thought they were getting ransomed for a second

AI 公司动态，非安全。

RT @thsottiaux: What I wanted to say yesterday is that we hit 25M active users and to celebrate we have now reset usage for all paid subscriptions for ChatGPT Work and Codex. See you soon for more news from The Reset Company.

恶意软件开发，但无新细节。

just added jabber noft on new infections for my c2/rootkit so it sends out the info to all users in the jabber list.

无实质内容。

@moyix hey maybe its just computer tho

编程话题，非安全。

I had heard that HashMap was slow, but oof https://t.co/tlFnO9EYHc

编程话题，非安全。

It's interesting when O(n^2) turns out faster O(n) because O(n) relies on a slow hashmap algorithm

硬件技术，非安全。

RT @jukan05: SemiAnalysis: SK hynix and Micron have struggled to achieve the highest HBM4 speeds. In particular, SK hynix had to redesign its base die, delaying HBM4 shipments for NVIDIA’s Rubin platform. Samsung currently has the best HBM4 technology. Broadcom has traditionally relied heavily on Samsung for its HBM supply. Samsung’s underperformance in the HBM3E generation made this a disadvantage for Broadcom’s ASICs. However, now that Samsung has pulled ahead of its competitors with HBM4, the same sourcing relationship has become an advantage for Jalapeño.

编程话题，非安全。

Ruby, escapes and good memories... https://t.co/9oNHb773gA

AI 训练动态，非安全。

RT @jun_song: OpenAI buying Macs in bulk for RL means two things: 1. The compute shortage is serious. 2. Apple Silicon is actually solid for this workload.

无实质内容。

@0xTriboulet they’re right, it will give you the same meth recipe i got at my local truck stop

观点，非技术。

It’s cute that people think Microsoft pushing everyone to Entra is security first and not recurring revenue first.