推特安全流

钓鱼链接识别难度上升，防御与用户教育需相应调整。

🚨 𝗗𝗲𝗳𝗲𝗻𝗱𝗲𝗿 𝗔𝗹𝗲𝗿𝘁: 𝗡𝗲𝘄 𝗚𝗼𝗼𝗴𝗹𝗲 𝗥𝗲𝗱𝗶𝗿𝗲𝗰𝘁 𝗨𝗥𝗟𝘀 𝗙𝗼𝗿𝗺𝗮𝘁 ​ A recent change introduces google[.]com/goto?url=... redirects for some search results, replacing direct website links. The target address is encoded in a proprietary format rather than displayed as a readable URL, making it more difficult for users to identify and verify a link's destination before clicking. ​ Over the past two days, analysis of sandbox results from @anyrun_app has revealed that several of these google[.]com/goto redirect URLs have been identified as malicious. The redirects ultimately led to harmful destinations, demonstrating how attackers may abuse the new redirection mechanism to conceal malicious content behind seemingly legitimate Google URLs. https://t.co/qa3d0Y1OLB ​ #threathunting

SmuggleMyPayload - generates HTML pages that embed and reconstruct files client-side using JavaScript. The toolkit offers multiple HTML Smuggling techniques, payload encoding options, JavaScript variations, and pre-built HTML templates. https://t.co/AOzGNbo37W

RT @BallisKit: Learn how to Weaponize Adaptix C2 with MacroPack and ShellcodePack! -> EXE and DLL Sideloading examples -> LNK spoofing PDF, HTA, Clickonce, etc. -> EDR Evasion options #redteam https://t.co/qzGm5KHp4E

The PID Mutation lesson is added into the BYOVD module of the Evasion course https://t.co/BUDJxbzHyD Remember thats the GitHub repo for this technique: https://t.co/qXABjGesWZ https://t.co/CcrshAUznE

A foundational C library for building operationally credible offensive capabilities https://t.co/JTb9aIeZSX

RT @InsiderGeo: NEW: 🇩🇪 A drone has crashed in Germany, just around 1 kilometre from a Bundeswehr military air base in Wunstorf, Lower Saxony, Wunstorf Air Base is the home of the German Air Force’s 62nd Air Transport Wing (LTG 62), which operates the Bundeswehr’s Airbus A400M transport fleet. It is the main base for Germany’s A400Ms and serves as a major hub for military air transport, including logistics, troop transport, MEDEVAC and aerial refuelling missions. Police are conducting a major operation after debris from the drone was discovered east of Lake Steinhude. Officers are searching the area for additional wreckage. The drone came down in an open, uninhabited area, and authorities currently say there are no indications of a danger to the public. However, the origin of the drone remains unknown. Investigators are working to recover and analyse more debris to determine what type of drone it was.

RT @interesting_aIl: Indian police dismantled a network that created over 500,000 fake Gmail accounts, each bypassing Google's own two-factor authentication, used to send hoax bomb threats to government offices Genuinely wild that fake accounts at this volume slipped past 2FA entirely https://t.co/9vMekl0f5j

RT @GoogleVRP: 📢📢📢 Attention bug hunters! Want to know more about how brutecat found a vulnerability in Google’s internal APIs, bypassing authorization to exploit the GFile library to gain access to internal filesystems and storage? If yes, check out his blog post 👇 https://t.co/ehwrnqHJlX

RT @MSNightmare2000: We released an update to fix the issue after the new antimalware engine update https://t.co/nrGgk52947

The product has some nice features, but the core component is the datasets used to train the LLM to detect threats and eliminate false positives. Kudos to @HackingDave and @Binary_Defense for being among the first in CyberSecurity. The market still isn’t showing any meaningful innovation compared to the models already out there. ⤵️

RT @zodiacon: How do you run your own service inside Svchost.exe? https://t.co/CPUiORW52b https://t.co/E1MfSluGAO

RT @HyperDbg: It's been a while since we passed 4K GitHub stars ⭐️ And more good news: HyperDbg's kernel-mode drivers now build on #Linux! 🐧 After successfully getting the user-mode components compiled, thanks to @MaxRaulea, we're now moving on to... https://t.co/kDf9BtRYCw https://t.co/ZgNO4RIk7N

https://t.co/vgseSsb4Sw

RT @LxlxIxlxlxL: i recently did some reversing on eac to emulate it and after breaking down their sub contextualized memory synchronization layers i found that their cr3 pivot swaps at 0xfffff8023a5de7f0 were completely vulnerable to page walk misalignment desyncs allowing for full thread deserialization within a self referencing enclave buffer this meant that by hyper extending the idt remap via a segmented ept hook at 0xfffff80764bb1290 i could seamlessly mirror syscall return states into an orphaned pml4 table effectively circumventing their heuristic based dispatcher validation eac s syscall gating handler sgh at 0x7ff67892f1d0 completely fails when you invoke a context aware tlb cache stall forcing their ntqueryvirtualmemoryinvocation layer into a false positive loop rendering their entire integrity verification stuck in a deferred stack traversal exception by artificially injecting a context switching null descriptor into their dr7 register offsets i was able to overload their event queue stack making eac believe it was still running while actually being fully desynchronized from the windows object manager to permanently halt eac service execution i hooked their obfuscated watchdog timer routine eac watchdog proc at 0xfffff80a37d45e90 and inverted its call stack through a synthetic race condition which forced their thread context to misalign its service dispatch table this caused their kernel resident page scanner to unmap itself leaving the entire anti cheat infrastructure frozen in a self referential memory loop by force injecting a recursive ntsuccess 0xc0000420 handler into their apc queue the entire system collapses into a segmented pte reallocation deadlock ensuring eac never recovers at this point the entire anti cheat exists in a suspended quantum state where their thread scheduler no longer acknowledges execution but continues reporting uptime to the watchdog service eac is finished

https://t.co/dt1wn4kCT6

RT @oneandonlyhusam: i recently did some reversing on eac to emulate it and after breaking down their sub contextualized memory synchronization layers i found that their cr3 pivot swaps at 0xfffff8023a5de7f0 were completely vulnerable to page walk misalignment desyncs allowing for full thread deserialization within a self referencing enclave buffer this meant that by hyper extending the idt remap via a segmented ept hook at 0xfffff80764bb1290 i could seamlessly mirror syscall return states into an orphaned pml4 table effectively circumventing their heuristic based dispatcher validation eac s syscall gating handler sgh at 0x7ff67892f1d0 completely fails when you invoke a context aware tlb cache stall forcing their *ntqueryvirtualmemory *invocation layer into a false positive loop rendering their entire integrity verification stuck in a deferred stack traversal exception by artificially injecting a context switching null descriptor into their dr7 register offsets i was able to overload their event queue stack making eac believe it was still running while actually being fully desynchronized from the windows object manager to permanently halt eac service execution i hooked their obfuscated watchdog timer routine eac watchdog proc at 0xfffff80a37d45e90 and inverted its call stack through a synthetic race condition which forced their thread context to misalign its service dispatch table this caused their kernel resident page scanner to unmap itself leaving the entire anti cheat infrastructure frozen in a self referential memory loop by force injecting a recursive ntsuccess 0xc0000420 handler into their apc queue the entire system collapses into a segmented pte reallocation deadlock ensuring eac never recovers at this point the entire anti cheat exists in a suspended quantum state where their thread scheduler no longer acknowledges execution but continues reporting uptime to the watchdog service eac is finished

https://t.co/tIR6HZ6jPL

RT @BourAbdelhadi: You don’t need a kernel 0day https://t.co/PkxkKlCQfk

And protection will only improve when Apple makes the `es_event_paste_t` Endpoint Security event available to 3rd-party security tools!! 🤞🏼 #SharingIsCaring See: https://t.co/hoWodAXxdJ

RT @vxunderground: For several months now I've been slowly, but surely, working behind the scenes to dramatically enhance vx-underground. I still don't have a definitive timeline, but I would like to share some stuff coming. 1. All source code will be moved from GitHub to vx-underground. The new vx-underground will have functionality to search through code bases by keywords for research, filter by language type, yada yada yada. 2. Enhanced malware paper searching. It'll be easier to search for stuff now. 3. API access to verified individuals. This is designed for researchers, or organizations, who may want the ability to perform large downloads programmatically. 4. Paid tier. vx-underground will remain free for everyone. However, if you're an organization which makes more than $1,000,000/year you will have to pay for some functionality (malware sample API downloads). I will not charge thousands of dollars, I'm not greedy, but some of you large companies are jerks and need to pay up for scraping the site for AI and samples. API access will be free for individual researchers, small businesses, students, non-profits, and government institutions. Unfortunately, I've learned the hard way large for-profit organizations will not donate. 5. I've heard your complaints for YEARS. We are looking for a HTML-only vx-underground implementation, probably as a subdomain or something. I'm well aware you hardcore nerds hate JavaScript. I will continue investigating this. 6. We're working on our 4th book, Black Mass Volume IV. I have more to announce, but some details are still up in the air. These changes likely won't appear until ... I don't know, bro. I've got a full time job and a family. I'm thinking probably late 2026, early 2027. Thank you to our donors and sponsors that make this possible. I'm not as speedy as I used to be with updates, but I'm still cooking. Cheers -smelly

Next two actions: 1) Kill https://t.co/BcSDPctaHH forever 2) Build a YouTube channel to visualize Purple Team Techniques To everyone who visited my first blog, thank you for reading. It’s time to put it in sleep mode. 🙏

RT @offbyoneconf: 🔥 DAY 2 IS OFFICIALLY ON! @nicowaisman kicked off Day 2 #OB12026 with an incredible journey from the early days of offensive AI to where we are today with his keynote “The Chaos Phase or How I Learned to Stop Worrying and Love Offensive AI” We’re bringing the house down today! https://t.co/obLniuAHxa

RT @offbyoneconf: @ThomasKing2014 brought serious research & even a few yoga poses to the stage #𝐎𝐁𝟏𝟐𝟎𝟐𝟔, to deliver his live demo to us at #OB12026 “Rooting the Google Pixel 10 with Old-School Techniques” 🎥 Massive shoutout to our l33t AV team for keeping up and making the demo shine! https://t.co/riRe4fdy4m

RT @offbyoneconf: 🕶️ BETTER THAN JAMES BOND! @freddo_1337 & @kiddo_pwn took us behind the scenes with their engrossing presentation: 🎯 “No Time to Pwn2Own: Exploiting After the Draw” Real-world experience, serious exploitation, a story that kept the audience glued to the #𝐎𝐁𝟏𝟐𝟎𝟐𝟔 stage! https://t.co/FTQmL48rYE

RT @ASDGovAu: We’ve updated our guidance on Detecting and mitigating Active Directory compromises. Learn how malicious actors target Active Directory and the steps organisations can take to identify risks and strengthen security. Learn more 👉 https://t.co/wvUlZes3T5 https://t.co/366eXLxlJT

RT @Enno_Insinuator: Part II: Sniffing Bluetooth Auracast, from @ttdennis via @Insinuator https://t.co/n0aeRy05Rh

RT @Unit42_Intel: A compromised regional news outlet’s WordPress site is using malicious JS to serve #ClickFix prompts, ultimately delivering a ConfuserEx-encrypted #StealC payload. One observed infection chain routed a targeted user via a local library's proxy. Details at: https://t.co/6lWAPK6R6L https://t.co/b8NGtsUbhw

RT @anyrun_app: 🚨 #PhantomStealer is a MaaS infostealer spread via business-themed phishing lures. It uses JS, WMI, hidden PowerShell, and in-memory execution to steal browser data, creating risks of account takeover and data theft. Full analysis and detection tips: https://t.co/dpddv1hpYk https://t.co/mglLGMtOSV

RT @RexorVc0: 🚨 New Research: VectraRAT 🐀 We at STRU have identified a previously undocumented #RAT operating under the #MaaS model. The dev has been active for years. This isn't their first build. Previously known as #Nyxel. VectraRAT combines #RAT capabilities with #Stealer functionality, managed through a centralized panel to control multiple victims simultaneously. Active campaigns confirmed. Victims identified worldwide and #USA among the top targets. 👽 Blogpost: https://t.co/KDopaOuuOq 🔗 Full report: https://t.co/iFpKliBjpQ #ThreatIntelligence #CTI #ThreatHunting #Malware #Threat #TA #Campaign

RT @JenevaFMRadio: Special operation of lost parcel claming. https://t.co/gDlDTBr6rn

We've just published the results of the Q3 2026 VBSpam test. The threats observed during this period demonstrate how modern phishing campaigns increasingly blend familiar business and consumer lures with lightweight evasive infrastructure. https://t.co/wd1nKUPj9V https://t.co/N65aMR9E7w

Acronis TRU uncovered a multinational campaign in which Red Heron, a Chinese-speaking threat actor, rapidly weaponized CVE-2026-60004 to compromise internet-facing instances of Gitea, a self-hosted source-code management platform. https://t.co/o1iDpGjDyL https://t.co/bPeLGNcWTx

RT @objective_see: Stoked that Shieldlink AI (https://t.co/piEjQMozMZ) has joined our "Friends of Objective-See" program 😍 Their support helps make possible our: 🛠️ open-source tools 📚 free #TAOMM books 👨‍👩‍👦 community-driven #OBTS/#OFTW events ...and much more! Mahalo Shieldlink AI 🙏🏽

RT @leonjza: New gowitness, 3.20! https://t.co/ZQ9DUkawrY https://t.co/lXNw7hiE88

falco v0.45.0-rc1 — Real-time cloud-native runtime security agent for Linux that monitors syscalls and container/Kubernetes metadata to detect... https://t.co/RES6Sw6Ffc https://t.co/12yvL2507l

Using Reflective Loaders to Replace LoadLibrary for Hot-Swappable Modules in C++ https://t.co/T5ZG4Z0kRE

ReflectivePluginLoader - A minimal PE mapper that loads DLLs straight from memory and calls into a clean plugin interface, no LoadLibrary needed https://t.co/nLWVLzD4Fo

https://t.co/lDAVn1TucV next one from my blog #macos #malware #persistence series https://t.co/Cpx1rtYU6U #hacking #malwaredevelopment #redteam #blueteam #cybersecurity #threatintel #research https://t.co/MAYRhQ8zO3

Research Worth Reading - Week 37, 2026: Attackers taking notes, garbage hidden in JSON, sandbox escapes and Ruby back in the news for the wrong reasons. https://t.co/jlZD0lrgYr

opennhp v1.0.1 — A lightweight, cryptography-powered, open-source toolkit built to enforce Zero Trust security for infrastructure, applications, and data... https://t.co/yJS3BMsKX4 https://t.co/SmpAKu0uuH

ffuf v2.3.0 — Fast web fuzzer written in Go https://t.co/Tih55Kou7J https://t.co/wKcRSClsHZ

Writing Beacon Object Files In Mythic Using Dark Agent For MacOS https://t.co/75M9L6Ttuv

tailscale v1.102.4 — The easiest, most secure way to use WireGuard and 2FA. https://t.co/kBxNBo5Jfj https://t.co/qWix8oo05x

Offensive SIEM - Practical techniques & queries for leveraging SIEM as an offensive discovery tool https://t.co/9NpVsIQ4VW

I yapped about this in a video without techno trance music. Also shrunk down the proof of concept to PowerShell. clicky clicky ---------> https://t.co/8O7Ra0Xe1w

firecracker v1.17.0 — Secure and fast microVMs for serverless computing. https://t.co/L3IsR4vNXZ https://t.co/hqdNVYCngo

These scores are insanely good. This is because of architectural changes from chipset to chipset or is it because of the control on entire environment from chipset to OS? Probably both?

RT @kevin_klyman: METR is just one of many existing AI evaluators! If you take issue with METR, that isn’t a good argument against requiring companies to carry out embedded audits Here are 20 other evaluators who work with AI labs to assess risk: Transluce Grey Swan Apollo AVERI SecureBio Faculty Vaultis Dreadnode Irregular RAND Far AI ActiveFence (now Alice) Active Site Deloitte (via Gryphon acquisition) Nemesys Mercor (via Sepal acquisition) AE Studio Scale Frontier Design Redwood

RT @CaidoIO: Two years ago, we asked what Caido was missing, and here we go again. What tasks still force you to switch to another tool ⁉️ ⁉️ Blunt is fine. We’re taking notes. https://t.co/tQfKI2itat

fleet fleet-v4.91.1 — Open-source platform to secure and manage endpoints via MDM, patch management, software deployment, and osquery-powered visibility with... https://t.co/0EmCeYiRId https://t.co/lSBTCWZFEk

RT @IntCyberDigest: ❗️ IP intelligence company Spur pushed a PR containing comments from their AI saying they'd copied competitor Synthient's CLI. Spur raised $200 million this year. Synthient is just one guy named Ben. https://t.co/dAKHtgLMDi

that’s what i thought

Woo!!

RT @bl4sty: I showed you my face TWICE, i uploaded my government issued drivers license (front and back); what more do you want man? there was no follow up requesting additional information whatsoever. this is a humiliating and utmost frustrating experience @sama >:-[ https://t.co/ekL7GuMKPu

RT @saaaanjjjuuu: My iPhone 15 Pro Max is stuck on bootloop after iOS 27 stable update 🤡 https://t.co/T63kvFi0Zr

@S1r1u5_ @sudhanshur705 is the goat!

RT @iM4CH3T3: 😎 https://t.co/tB3bOSEZ37

RT @MiaAI_lab: Ash is building a direct link between DGX Sparks and a Mac Studio. He's got Metal and CUDA moving data over real RDMA. The idea: let the Sparks handle the heavy prompt processing while the Studio generates tokens - so together they could outperform either setup on its own. That's 256 GB on the Studio + 128 GB per DGX Spark = 512 GB total. A ~400 GB model wouldn't fit on either system alone, but it could potentially run across all three if the split works. Seriously impressive stuff!

RT @MiaAI_lab: https://t.co/MEWmnkACZa

RT @S1r1u5_: i hope this is true, but i am quite pessimistic about it. intuitively it still seems offence dominant to me. there is already a huge amount of attack surface from the vast amount of software and code we have, and now agents are adding even more while becoming part of the attack surface themselves. i am also pessimistic about formal verification for various reasons. software intent is not clear unlike math, huge amount of state space with interconnected component etc and software bugs are not even the only way to attack things. you have social engineering, and now prompt injection etc. claims like “we found all the p0s” seem absurd and wrong threat model to be working with. i generally think a good way to deal with cyber threats is to just assume everything will eventually get pwned and build things around that assumption. reduce the scope, isolate things, limit privileges, etc,

RT @halvarflake: This is worth reading, as it is reasonable and well-argued. I agree on many points and disagree on some, but it's a text that can be taken seriously.

@cr3ghost The horses seem to like it

@_RastaMouse I’ve never used the grass for a quick one, but not a bad idea.

@sudhanshur705 https://t.co/xjtZcEfLHr

Finally some serious LLM research

RT @Laughing_Mantis: Since I discussed my concerns on cybersecurity and AI, I have seen so many people start discussing how we are not at the table and asking for us as a community to be involved. Thank you, please continue to be loud and vocal. We are building a team to help

RT @MiaAI_lab: Leaning towards 4x DGX Sparks bcs I expect prefill and concurrency performance to be better on Nvidia stack even vs the new M5 Ultra. The advantage of a Mac is it's one unit vs 4. Decode speed should be faster on a Mac, but with TP=4 on the sparks, the gap may be smaller than you’d expect

RT @MiaAI_lab: DeepSeek v4.1 Flash through API costs less than $3 for over 500 Million tokens. Even more crazy: I measured a peak of ~315 tok/s decode and ~52k prefill

RT @thdxr: people do not understand how big the ai market is everyone can win well not you. but everyone else can

🎉🥳🎂 https://t.co/Q60uIvaa92

and my thoughts on formal verification shared in below tweet. my hunch is we underestimate the complexity of software, like we can only understand what we have the ability to, i feel same applies to ai. i feel like every new generation of ai opens up more of the vulnerability space that was always there but we simply its not intelligent enough to find. so i think the safer bet is to assume you will eventually get pwned rather than rely on the assumption that you have exhausted the vulnerability space. especially when the attacker has incentive to keep throwing increasingly intelligent ai at finding the thing you missed.

yea on defence dominance, i am probably underestimating it, if continuously running ai can find bugs and also write a patch without breaking and introducing new bugs(which is the main bottleneck) then ig software would get secure. this requires a world where software entirely run by ai without human middleman otherwise it will be just slow, amdahl law.

RT @xeophon: @jphme @tenobrus I am a huge proponent of the time lag between open and closed models. But then you look at what cybersec professionals are dealing with and see that they have to beg closed providers and get ignored. Awful!

It's always Norway 🇳🇴 https://t.co/EMddaUVC8n

https://t.co/aJMl5Ji1Ke

If you have a profile on BlueSky and you are already following me here for content, you might want to follow me there as well to avoid losing content due to X restrictions on visibility. Link to my profile in the reply ⤵️ https://t.co/WWikEchl3O

the research sudi does and pwning he pulls off these days is crazy, i will make sure he writes about them xD

RT @sudobunni: Hellooooo I have a favor to ask... I'm trying to get a streamer award this year and I've just submitted my own application under the categories: Sapphire Award, Rising Star Award, League of Their Own, and Best Creative Arts Streamer. It would be hugely impactful for me to be a nominee or winner! It closes on October 2nd, so please get your nominations in before then. Thank you thank you thank you :) https://t.co/2k6qVFnlKs These category numbers are 2, 17,33, 35,37 I would be the first female coding creator to be nominated at these awards!

i’m proposing something that can both help us slow down a bit and be less doomer. WoW Forever horde guild full of schizo hackers and AI researchers with psychosis. its like TPOT basement. weird rustaceans can come too i guess

wtf David Sacks has been awesome

RT @CBSNews: EXCLUSIVE: Trump AI advisor David Sacks tells CBS News' @JoLingKent that tech leaders like Anthropic CEO and co-founder Dario Amodei have a responsibility to ensure their products are safe — and should "step aside" if they can't control what they're building. https://t.co/mDq88h7pCA

ill continue to trash these companies while (ab)using their models to get paid. that is accurate. you should too. you’re not changing the outcome. but you should use AI to your advantage while it unfolds. just use AI and play WoW Forever. its the best time ever, don’t doomer

if you still think it was all just a big whoopsy, i got news for you

RT @dwarkesh_sp: Was really interesting to hear John, Beren, and Charlie speculate about why Sonnet 5 and Opus 5 feel like worse models than GLM 5.3 (despite the fact that Anthropic can do raw logit distillation from Fable, and can also train Sonnet/Opus on the environments from which Fable was trained). Led to some interesting thoughts about value of distillation, what it takes to do distillation effectively, and what kinds of model behaviors are hard to extract from distillation.

RT @trapdoor456: That Popeye's Fried Chicken is damn good! 👀👀 @Popeyes https://t.co/qKQOXTvvO8

RT @jenniferzeng97: Viral claim this week: Huawei founder Ren Zhengfei and his family went dark after a 1 a.m. van left for the Dapeng coast. Police video at HQ. Rooftop wage protest the day before. Huawei has not said a word. Taiwan media ran the rumor. The company’s U.S. criminal trial opened in Brooklyn last week. Unverified. Full report in the video. https://t.co/dd9gJoOoRU

RT @IceSolst: bro what https://t.co/hozwLmN17r

RT @136Division: The accent of western Liaoning (area link btwn northeast & north China) is like this— they tend to make the tone rise at the end, so it always sounds like they're questioning you, or comes off as really provocative. Many ppl who don't know this think the locals are unfriendly. https://t.co/r5zj9vcXBk

RT @CyberTruckFan9: @dnystedt If true, MASSIVE

RT @dnystedt: Rumor: Huawei founder and CEO Ren Zhengfei (任正非) and his family have fled China ahead of new regulations that begin Sept. 15 and would restrict entry and exit to China, media report, adding some reports allege he escaped China by sea from the Dapeng Peninsula. Others say there is a heavy police presence at Huawei’s headquarters. Huawei has so far made no official statement. 1/3 #huawei

RT @MiaAI_lab: Yeah I see why Dario is afraid Based on my initial testings DeepSeek v4.1 Flash is outperforming Opus 5 and is very close to Fable 5.1 across the board in frontend web design. And it's running locally on my little sparks!

RT @justdionysus: I’m really excited to dump out my brain this Saturday. I’ll be trying to explain my view of exploit development, mitigations, and how the process creates an interesting challenge for LLMs (and humans!) using some examples from our recent XNU research. Please say hi if you attend.

RT @nexta_tv: Chaos at BRICS: participants struggled to line up for the group photo Putin was clearly trying to make his way toward the center of the frame and looked noticeably uneasy at the prospect of ending up next to the Iranian president. He did not even spare his close ally a glance. In the end, the strategic spot in the center was secured — but not without some drama. Modi pulled Putin by the hand, causing him to stumble and nearly lose his balance.

RT @Polymarket: JUST IN: Nvidia CEO Jensen Huang declares AI doomer predictions are “made up.”

RT @visegrad24: BREAKING: A Russian Navy frigate has fired flares at a Danish military helicopter near Gedser, Denmark https://t.co/H1BS23YFxL

RT @HackingLZ: I’m a fan of people discussing offensive capabilities and what models can now do. At the same time, the people getting the best results from models have extremely strong and talented teams behind them. AI raised the floor, but many people didn’t know where the floor already was because they weren’t reading CTI, DFIR reports (@TheDFIRReport ), attacker playbooks, or leaked attacker chat logs. A lot of “real” intrusions were never especially sophisticated. Many organizations have meaningfully improved their security over the years. Others were forced into the game by ransomware, which turned otherwise junk intrusions with little payoff into multimillion dollar businesses. If ransomware and extortion disappeared tomorrow, plenty of smaller organizations would still have compliance and good faith reasons to invest in security, but not the existential fear driving them today. Machine speed is great and all, but ransomware groups still have to strike some balance between making money and becoming enough of a priority to get their doors kicked in. It’s also worth reminding people that the attacks causing the most sleepless nights so far this year have been supply chain compromises, not agents speed running commodity exploits.

RT @T3chFalcon: Proton surveyed over 7,000 mobile VPN apps available in the US. 85% of VPNs downloaded in the US contain trackers that collect and share your personal data. your device ID. your network type. your mobile carrier. your usage patterns. all sent to third parties. 64 of those apps are owned by Chinese companies. under Chinese law, companies must share user data with the government on request. 31 of those 64 Chinese-owned VPNs hide their real ownership behind shell companies in Singapore, Hong Kong, and the UK. 64 VPN apps actively track your physical location using GPS. the tool you installed to protect your privacy is tracking where you physically are. some of those are among the most downloaded VPNs in the US — VPN Proxy Master, X-VPN, VPN-Fast. trackers in US VPN apps send data to: — China: 1.5 million downloads a month — Russia: 1.4 million downloads a month — Israel: 2.6 million downloads a month Apple and Google let all of this into their app stores. they check that a privacy policy exists. they don't audit whether the app actually follows it. What to look for in a trustworthy VPN: — independently audited no-logs policy — open-source code — avoid free options Proton VPN (open source, Independently audited — (https://t.co/eqhbG0Tnnm). the VPN that tracks where you go is worse than no VPN at all.

RT @visegrad24: BREAKING: 🇵🇱 A photo of the drone off Poland’s Baltic coast has been released "More and more data indicates that the Russians likely lost a new model of Gerbera-type drone," said Poland's Defence Minister Władysław Kosiniak-Kamysz https://t.co/vsypFGXXaO

Yes. AI can be applied productively to cyber defense. Sometimes people make statement like "cybersecurity is offense dominant" but we need to also take into account that finding vulnerabilities can help patch them more rapidly.

RT @S1r1u5_: i hate security disclosure so much. the number of companies that take our work for granted is insane. how about you talk to us collaboratively from the start instead of being asinine about it? like, give us some respect, bro. we worked on this shit for weeks/months and showed you something that could have become a massive disaster if a bad guy finds it, and you don’t even seem to give a fuck and see us like some villains?

more on this https://t.co/dRgd9l5DZP

the bug me and @sudhanshur705 was exploit at that time is this. on basecamp. https://t.co/rJmSvjia9i

this was me few years back, the pain of approaching a new target and writing an n-day exploit used to be a lot. i would spend days just understanding all the new things the target had implemented(like the v8 sandbox, windows exp here)then more days figuring out everything else and improvising around it. it was literal grunt work, but it was beautiful, time would just disappear, you had a clear goal, you got feedback at every step, and you kept hillclimbing until you got there. i had some of the most fun in my life doing this kind of work. unfortunately, that is changing now, i don't need to spend my time on this, this is all just largely automated.

RT @Guluisacat: iOS 27 App Sandbox Escape Photos... App Containers Submitted as a low impact bug a few days ago, but after a closer review today escalated it to an app sandbox escape https://t.co/fuDGYLP61v

RT @hrkrshnn: If you're a security researcher and you think there are bugs or exploits that are out of reach for agents today, can you DM me?

Dump Local Credentials with VSS New Medium post. Today we are going to use the Volume Shadow Copy Service API in C++ to read files that Windows keeps locked while the system is running https://t.co/Gx1DxILulp

RT @nmatt0: I feel like this was exactly why I got into local AI now.

RT @codeneverdies: Local privilege escalation vulnerability in steam client's service, the video below showcases launching a system shell by exploiting their install script functionality. https://t.co/tynsWvqZZU "BrokenPipe" by @killa

RT @zux0x3a: Just published a new research on turning WerEnc.dll newly discovered LOLBin into a native encryption primitive with BYOK. The write-up and PoC are released 📁 https://t.co/waO6llVtoc

RT @tjbecker: Did OpenAI just change their threshold for cyber classification? just started tripping this in multiple different contexts that were fine a few hours ago. Hearing the same from some friends https://t.co/x0qYs63LcG

RT @anand_himanshu: I set out to relearn Linux memory management. I ended up controlling a page table and capturing a Google kernelCTF flag. Known bug, lots of wrong turns. Here's the write-up: https://t.co/H58niPCakC

RT @Blackfrost_AI: Y'all thought I was lying. Backup Your models people. Stay Frosty https://t.co/Z6BsyPWd7A

i wasnt paranoid when i said the huggingface hack was on purpose. just early.

RT @senzee1984: @malware_owl Thank you for liking it ☺️ The English version of it for more convenient reading https://t.co/EXBVh4b8NE

garak v0.17.0 — Modular LLM vulnerability scanner that probes for hallucination, data leakage, prompt injection, jailbreaks, and toxicity using static,... https://t.co/yiW7M6VlNs https://t.co/vlZ4x2QB02

RT @andrd3v: Missed one more from the drop lol CVE-2026-43783 - macOS LPE via DesktopServicesHelper. One XPC request -> arbitrary chown -> root. PT SWARM writeup: https://t.co/uEaHXKbLhb My blog: https://t.co/TlMOqux5lW PoC: https://t.co/qH1LnxXbjj