Going Around EDR – PSW #900

Paul Asadoorian

1. Repurposing Dodgy Android TV Boxes As Linux Boxes

   I've done this, it's interesting, Debian is not the greatest experience for a TV box (often you'd just use a browser to navigate to Netflix to watch). One thing that struck me: Android TV boxes advertise higher CPU/RAM/Storage than what they are actually selling. Go figure, I mean they include malware AND false advertising, yet Amazon sells tons of them, yet they won't sell a Flipper Zero. WTH?

2. Using Ghidra to patch my keyboard’s firmware

   Sometimes reverse engineering is not looking for vulnerabilities, but modifying firmware to add or change functionality. This is an awesome example.

3. The YouTube Ghost Network: How Check Point Research Helped Take Down 3,000 Malicious Videos Spreading Malware – Check Point Blog

   "The operation relied on cracked software and game hack videos to lure victims into downloading password-protected archives containing malware." - People want cracked software and game cheats, so attackers used YouTube to trick people into downloading fake cracks/cheats. Sounds familiar? I am pretty sure this technique dates back to the days of BBS.

4. Invasion of the Face changers: Halloween Hijinks with Bluetooth LED…

   Love this hack so much, I want to test it myself. I have one of these masks, but I don't believe it is the Bluetooth version. With BT, an attacker can upload new images and tell the mask to display that image. The Bishopfox team uploaded their logo, but your imagination can now run wild as to which image(s) you want to display on another person's mask.

5. codykociemba/NoLongerEvil-Thermostat: Breathe fresh life into your bricked Nest Gen 1 & 2, now with 100% less evil!

   I can't express how much work this is and how amazing this project is, check out the high level: "The custom firmware flashes the device with modified bootloader and kernel components that redirect all network traffic from the original Nest/Google servers to a server we specify. This server hosts a reverse-engineered replica of their API, allowing the thermostat to function independently while giving you complete control over your device data and settings." - However, it would be nice if you could host the server yourself, maybe you can, but a quick scan of the docs shows that you have to register with https://nolongerevil.com/

6. Tailscale Peer Relays: High-throughput relays for secure, flexible networks

   Tailscale Peer Relays is a new feature made publicly available in beta, offering a customer-managed traffic relaying option alternative to Tailscale’s managed DERP servers. This allows Tailscale nodes to relay traffic for peers in the same tailnet, including for themselves, using a high-throughput and low-latency relay based on UDP, embedded directly in the Tailscale client. It is designed to improve performance where direct connections are not possible, such as behind strict firewalls or in cloud environments with NATs.

   Peer Relays are managed by the customer, providing throughput comparable to direct connections, often much higher than DERP relays. They enable connections in "hard NAT" and firewall-restricted environments by only requiring a single open UDP port per relay. Available on all Tailscale plans, including free, with two peer relays allowed initially. The feature is still in beta, with ongoing improvements in connectivity and debugging. "A Tailscale DERP (Designated Encrypted Relay for Packets) is a relay server used by Tailscale to facilitate device-to-device connections—especially when a direct connection between devices in your Tailnet cannot be established due to strict NAT, firewalls, or other network barriers."

7. Epic Pentest Fail – SpecterOps

   We've all been there, on a pen test, and made an oops. Good for Spectre Ops for sharing their lessons learned. Most pen test posts are like "look at all the cool stuff I did" rather than "we made an oops, learn from it".

8. Dead Domain Discovery: Discover Expired or Unregistered Domains

   Neat tool for finding dead domains.

9. Demonstrating a Rollback Attack on a Honda via HackRF Portapack and an Aftermarket Security Solution

   I need to test this: "Over on YouTube "Obsessive Vehicle Security" has uploaded a video demonstrating a rollback attack against a Honda vehicle using a HackRF Portapack and the "Remote" function on the Mayhem firmware. His recent blog post also succinctly explains the various types of keyless vehicle theft used by modern thieves, including Roll-Jam, Relay Amplification and Rollback attacks." - Now I just need an older Honda :)

10. EDR-Redir V2: Blind EDR With Fake Program Files

    Moar EDR bypass: "EDR-Redir V2 is a technique that leverages Windows bind link technology to redirect the operating folders of Antivirus and EDR solutions, such as Windows Defender, through a controlled folder and disrupt their ability to monitor or protect their core files. Traditionally, EDRs block file writes in their own directories, but they cannot block actions in their parent folders (like "Program Files"), as doing so would impact overall software functionality. By creating strategic bind links—redirecting folders to themselves and excluding the EDR’s own subfolder—it's possible to break the security model, tricking the EDR into seeing a different folder as its parent and enabling DLL hijacking opportunities"

11. SonicWall: Firewall Management Lessons After Breach Incident

    This is interesting, Sonicwall and F5 have both been breached, and I believe Cisco had a breach earlier this year as well. Threat actors are not just going after network device vendors?

    • Note:The link was 404, I fixed it --Sam
12. YouTube Goes Bonkers, Removes Windows 11 Bypass Tutorials, Claims ‘Risk of Physical Harm’

    Another story of content being removed, then put back, and around we go. This time someone explaining how to get around Windows 11 requirements gets flagged for "risk of physical harm". Funny, but also not funny.

13. Data breach at Chinese infosec firm reveals weapons arsenal

    Interesting: "Chinese infosec blog MXRN last week reported a data breach at a security company called Knownsec that has ties to Beijing and Chinas military. MXRN says the company leaked over 12,000 classified documents, “including information on Chinese state-owned cyber weapons, internal tools, and global target lists.” The trove also apparently included evidence of Remote Access Trojans that can crack Linux, Windows, macOS, iOS, and Android. The Android code can reportedly extract information from popular Chinese messaging apps, and from Telegram. Also in the haul was a spreadsheet that lists 80 overseas targets Knownsec has successfully attacked, plus 95GB of immigration data obtained from India, 3TB of call records stolen from South Korean telecom operator LG U Plus, and 459GB of road planning data obtained from Taiwan. The blog says attackers posted some of the documents to GitHub, which quickly removed them."

14. Threat Landscape of the Building and Construction Sector: Initial Access, Supply Chain, and IoT
    • The construction and building sector in 2025 faces heightened cybersecurity risks due to increased digital transformation and dependence on IoT-enabled equipment, BIM systems, and cloud-based tools. Threat actors—including ransomware operators, organized crime networks, and APTs from countries like China, Russia, Iran, and North Korea—are aggressively targeting this industry for financial gain, espionage, and extortion.
    • Attackers exploit insecure IoT-heavy machinery, project management systems, and third-party vendor relationships, often via phishing, compromised credentials, software supply chain breaches, and social engineering.
    • Initial network access is sometimes bought on dark web forums, as access brokers sell stolen credentials for VPN, RDP, SSH, Citrix, etc., allowing lateral movement and data exfiltration once inside.1
15. Analysis of NGate malware campaign (NFC relay)

    How does PCI address this Jeff? - "NGate is a Android NFC relay kit used to cash out ATMs with victims’ own cards. It’s delivered via phishing plus a “bank support” call that pressures the user to install an app, tap the card to the phone, and enter the PIN. The app runs in reader mode to capture EMV APDUs and the PIN, then exfiltrates them via a simple framed TCP protocol to a hard-coded C2; the same family also ships a payment-category HCE service, enabling an emitter role at the ATM. Configuration is stored as an XOR-encrypted asset with a key derived from the APK signing cert (SHA-256), which in this sample resolves to a live, plaintext C2. Bottom line: once the card is tapped and the PIN is entered, the attacker can relay the session and withdraw cash."

16. CVE-2024-12649: vulnerability in the Canon TTF interpreter

    I so want to see this exploited in the wild, perhaps give people some motivation to harden and update printers:

    • The vulnerability CVE-2024-12649 affects Canon printers, specifically their TrueType Font (TTF) interpreter, allowing attackers to execute malicious code simply by sending an XPS file containing a specially crafted TTF font for printing. This attack exploits insecure virtual machine instructions in the printer's DryOS firmware, leading to a stack buffer overflow and code execution on the printer processor.
    • Attackers deliver a malicious XPS file (often via email or messenger) and convince an employee to print it.
    • The XPS file contains a font with code that targets vulnerabilities in the printer's font processing engine.
    • Successful exploitation allows attackers to run malware within the printer, potentially creating tunnels to pivot further into the organization's network or exfiltrate printed data.

    I love the opening statments: "These days, attackers probing an organization’s infrastructure rarely come across the luxury of a workstation without an EDR agent, so malicious actors are focusing on compromising servers, or various specialized devices connected to the network with fairly broad access privileges yet lacking EDR protection and often even logging capabilities. We’ve previously written in detail about the types of vulnerable office devices. Real-world attacks in 2025 are focused on network devices (such as VPN gateways, firewalls, and routers), video surveillance systems, and the servers themselves. But printers shouldn’t be overlooked either,"

17. Sniffing established BLE connections with HackRF One

    Lots of detail here, summary: "The article demonstrates that with custom firmware and scheduled frequency changes, the HackRF One can be converted into a powerful BLE analysis tool, overcoming major hurdles in passive sniffing and enabling advanced security research possibilities." - No public tool, have to request it.

18. All firmwares for Flipper Zero, comparision and help to choose – Awesome Flipper

    Here are the three best ones, I've tested all of them:

    • Unleashed - this firmware is based on the official firmware and is suitable for those who already know what they need and what the official firmware does not provide. Minimal changes in the interface, the emphasis is on functional and useful changes.
    • Momentum - custom firmware is based on the Official Firmware, and includes most of the awesome features from Unleashed. It is a direct continuation of the Xtreme firmware, built by the same (and only) developers who made that project special.
    • Roguemaster - firmware is based on Official firmware and distributed as a firmware combining Unleased and Xtreme firmware features with additional apps and animations.

    Note: The above comes directly from the Awesome Flipper site. I tend to bounce between the above three firmware options based on the features they offer at the moment. Unleashed has a daily build that I've tested, works well, and includes interesting features, such as vehicle keyfob cloning, which I have been wanting to test.

19. Curly COMrades: Evasion and Persistence via Hidden Hyper-V Virtual Machines

    "The attackers enabled the Hyper-V role on selected victim systems to deploy a minimalistic, Alpine Linux-based virtual machine. This hidden environment, with its lightweight footprint (only 120MB disk space and 256MB memory), hosted their custom reverse shell, CurlyShell, and a reverse proxy, CurlCat. By isolating the malware and its execution environment within a VM, the attackers effectively bypassed many traditional host-based EDR detections. EDR needs to be complemented by host-based network inspection to detect C2 traffic escaping the VM, and proactive hardening tools to restrict the initial abuse of native system binaries." - This is a neat technique, certainly not new, but perhaps now more widespread. Monitoring the network remains crucial.

20. 9 Malicious NuGet Packages Deliver Time-Delayed Destructive …

    This reminds me of Stuxnet:

    • The Sharp7Extend package is designed for ICS and SCADA deployments, where it's commonly used as a .NET library to integrate with Siemens S7 programmable logic controllers (PLCs), a widely deployed standard in industrial automation.​
    • In ICS settings, such libraries are essential for enabling communications between software applications and factory-floor PLCs, often governing critical processes in manufacturing, energy, and automation.​
    • The malware exploits extension methods (notably .BeginTran and .Exec) to intercept PLC and database operations transparently, making attacks nearly impossible to detect amidst normal industrial traffic.​
    • Malicious behavior in such environments could result in undiagnosed process failures, silent corruption of production data, and dangerous manipulation of safety-critical systems, with the potential for physical damage or operational shutdowns.​
    • ICS victims include organizations running .NET-based SCADA/HMI software or custom industrial automation tools that utilize these NuGet components for real-time PLC communications
21. Quantifying Swiss Cheese, the Bayesian Way

    Interesting to blend these two methods:

    • Swiss cheese: "Think of your organization’s defenses as layers of Swiss cheese. Each layer has some holes (no control is 100% effective), but multiple layers together can reduce the chance of an attacker slipping through all of them. This is the classic Swiss cheese model of risk management. If the holes (control failures) don’t line up, the threat is stopped; only when all holes align does the bad outcome occur."
    • Baysian Analysis: "This is where Bayesian inference enters the picture. Applying this to exploit likelihood means we can treat each control as a probabilistic filter that reduces the base exploitation likelihood from EPSSg. To do this rigorously, we need a way to quantify how effective each layer is — and to update that belief over time."
22. So You Want to Hack Back

    There is a table in this post that lists the "Hack Back" activities and indicates whether they are permissible. It's pretty accurate < That's my opinion.

Jeff Man

1. Nikkei’s Slack breach leaks sensitive data from more than 17,000 users

   Slack was breached but was it Slack's fault? Look for the expert commentary for insights into this question.

2. Polymarket Suffers Major Security Breach

   Phishing the comments section? This article cites an "urgent need for enhanced security frameworks". Because...nobody is following existing security frameworks?

3. For over 40 years, this woman kept a secret so classified that not even her husband knew…

   Saw this on Facebook and thought it was a good read. Enjoy.

4. Phishing Campaign Targets Cloudflare Pages and Zendesk to Mimic Support Portals

   Maybe the miscreants aren't sophisticated but they are clever. So that AI Support Bot you are arguing with might be hostile?

5. Zendesk Security Breach: Phishing Campaign Targets Customer Data

   Going after Cloudflare in a new (and creative?) way.

6. Phishing attacks succeed by exploiting human psychology, manipulating users by leveraging trust, familiarity and a sense of urgency to trick them into clicking links, entering credentials, or installing software.

   If a company takes more than nine hours to respond to an email breach, it is almost definitely going to also get a ransomware infection, experts have warned. tick tock.

7. You’re Running Out of Time to Claim Your Piece of AT&T’s $177M Security Breach Settlement
8. SonicWall blames state-sponsored hackers for September security breach

   related to Paul's story #11 - blame for SonicWall breach goes to state-sponsored hackers. (Paul: Updated my story number)

Lee Neely

1. Hardware hacker installs Minecraft server on a cheap smart lightbulb — single 192 MHz RISC-V core with 276KB of RAM, enough to run tiny 90K byte world

   A hardware hacker has installed a Minecraft server on a cheap smart lightbulb. Vimpo shows how this feat was completed and demonstrates the server working in a brief video. Key to this achievement was the bulb’s BL602 RISC-V-powered microcontroller.

2. Pentagon begins enforcing CMMC compliance, but readiness gaps remain

   This week marks the official rollout of the US Defense Department's Cybersecurity Maturity Model Certification (CMMC) program. The program establishes standards for how defense contractors handle controlled unclassified information. The plan establishes a tiered model of cybersecurity standard requirements for covered entities based on the sensitivity of the data they handle. The tiers will be introduced over a three-year period. Level one requires an annual self-assessment and affirmation of compliance with 15 core security requirements as described in the Federal Acquisition Regulation (FAR). Levels two and three involve assessments conducted by third-parties of compliance with additional security requirements..

   CMMC 2.0 is here. If you're a Defense contractor, make sure that you're already implementing needed controls, regardless of perceived gaps, this marks the beginning of phase one of a three-year implementation plan. Phase 1 allows self-attestation of compliance for CMMC Level 1 & 2.  Phase 2 starts November  2026 and requies a certified third party assessor (C3PAO),  Phase 3 begins Novembre 2027, and requires certificaton from the Defence Industrial Base Cybersecurity Assessment Center (DIBCAC). Note you can move to the external assessers sooner than those phases requrire, and you're going to need to make sure you're implementing needed controls, policies and processes.  Allow more time than you think you need, you don't want to be caught unpreparred and lose contracts.

3. Bank of England says JLR’s cyberattack damaged UK GDP growth

   According to the Bank of England (BoE), a debilitating cyberattack in August 2025 that caused a five-week shutdown of Jaguar Land Rover (JLR) operations is partly responsible for lowering UK headline GDP growth by a tenth of a percent compared to the prediction for Q3. The BoE's Monetary Policy Report follows the Cyber Monitoring Centre's assessment that the incident affected over 5,000 businesses and may be the "most economically damaging cyber event to hit the UK," with an estimated financial impact of £1.9 billion (about US$2.5 billion).

   The JLR attack is thought to be the first case where a cyber attack caused material economic and fiscal harm to the U.K. When calculating impact, make sure to include third-party impacts, suppliers where a significant part of their business depends on your viability. Your recovery may be dependant on their viability, make sure that is part of your planning.

4. QNAP Patches Multiple Vulnerabilities

   QNAP published multiple advisories to address vulnerabilities in their products, including seven critical flaws in the company's network-attached storage (NAS) devices that were found last month at Pwn2Own Ireland 2025. Critical vulnerabilities were addressed with updates for Hyper Data protector, HBS 3 Hybrid Backup Sync, QTS and QuTS hero, Malware Remover, and QuMagie. In addition, updates address important vulnerabilities in Download Station, Async Central, and QuMagie.

   It's been a minute since we've talked about NAS vulnerabilities. Don't get distracted by the flaws from the Pwn2Own event, QNAP fixed other vulnerabilities as well. This is a good time to make sure all your QNAP services are up-to-date, as well as verifying their security, removing unneeded/unauthorized accounts and services/apps. Verify you're subscribed to QNAP's security advisories as well.

5. Edtech company fined $5.1 million for poor data security practices leading to hack

   Illuminate Education, an education technology company, has reached a settlement with three US State Attorneys general over a 2021 breach that compromised student personal information, including medical conditions and special education accommodations. In December 2021, threat actors accessed an Illuminate online account using credentials of an individuals who was no longer employed by Illuminate and downloaded student data. According to a press release, "Illuminate failed to encrypt student data, implement appropriate systems and processes to monitor for suspicious activity, decommission inactive user accounts, and limit account permissions to only those that were necessary, ... failed to delete student data when its contracts with certain school districts ended and failed to conduct a complete investigation following the data breach." Illuminate will pay the states of California, Connecticut, and New York a total of US$5.1 million. The settlement also requires Illuminate to adopt a set of security practices that will better protect the data they hold.

   The lessons here should be clear, disable inactive accounts immediately, limit access to only what's required, encrypt sensitive data, and do a full investigation of any breach. Verify that you have these all covered, as well as MFA, EDR and monitoring. Where you have gaps, initiate a trackable project with a vetted timeline staff is accountable for to ensure resolution.

6. CBO systems accessed in ‘security incident’ possibly tied to foreign hackers

   The US Congressional Budget Office (CBO) has confirmed that it experienced a cybersecurity incident last week. According to a CBO spokesperson, "the Congressional Budget Office has identified the security incident, has taken immediate action to contain it, and has implemented additional monitoring and new security controls to further protect the agency’s systems going forward." The incident, which was first reported by the Washington Post, is under investigation. CBO was established in "1974 ... to provide objective, nonpartisan information to support the Congressional budget process and to help the Congress make effective budget and economic policy." CBO is a federal legislative branch civilian agency with a staff of 275. The agency's requested budget for 2026 reflects an 8 percent increase over the previous year; about half of that increase is earmarked for spending on enhanced cybersecurity and IT infrastructure.

7. Chrome 142 Update Patches High-Severity Flaws

   Google updated the Chrome stable channel for desktop to version 142.0.7444.134/.135 for Windows, version 142.0.7444.135 for macOS, and version 142.0.7444.134 for Linux to address three high-severity and two medium-severity vulnerabilities. The high severity flaws are an out of bounds write issue in WebGPU (CVE-2025-12725) and inappropriate implementation issues in Views (CVE-2025-12726) and V8 (CVE-2025-12727). The medium-severity flaws are both inappropriate implementation issues in Omnibox (CVE-2025-12728 & CVE-2025-12729). Users are urges to restart their browsers to ensure the updates are installed.

   It seems like we just went through this doesn't it? I am putting a reminder on my calender to check/restart browsers weekly, likely a Monday thing - start the week off fresh. There is some debate as to weathre the flaws categorized as medium are really critical, at this point the time spent debating the criticality versus just rolling out an updated browser isn't worth it, just do it. Where you're managing browsers, you may want to shorten the max interval users can defer the browser update. These days 48 hours is really too long. Don't forget to track updates on all your platforms, not just Win/Mac.

8. LANDFALL: New Commercial-Grade Android Spyware in Exploit Chain Targeting Samsung Devices

   Executive Summary Unit 42 researchers have uncovered a previously unknown Android spyware family, which we have named LANDFALL. To deliver the spyware, attackers exploited a zero-day vulnerability (CVE-2025-21042) in Samsung’s Android image processing library.

   CVE-2025-21042, out-of-bounds write in libimagecodec.quram.so has a CVSS score of 8.8. Make sure you're running libimagecodec.quram.so from Samsung's Maintenace Release in from April (SMR-April-2025) or later.  Note SMRs only include security patches, not bug fixes, so there is low risk in applying them. You can leverage Samsung's Knox E-FOTA (firmware-over-the-air) to deploy these updates without user interaction.

9. GlassWorm Returns: New Wave Strikes as We Expose Attacker Infrastructure

   Koi security has published a blog post noting a new wave of GlassWorm malware in the OpenVSX marketplace on November 6, 2025, despite access token rotation and preventive security measures taken by OpenVSX when the campaign was first detected in October. GlassWorm hides malicious code in VSCode extensions using invisible Unicode characters, using a Solana blockchain for command-and-control (C2) that downloads an infostealer and ultimately turns the infected system in to a persistently backdoored proxy server. Attackers can use the stolen credentials to compromise additional packages and extensions, continuing infection through the supply chain.

   Koi's blog includes IoCs for the November campaign, grab them and go hunting. The GitHub campaign seems to currently be limited to JavaScript projects. Alert developers about Unicode misuse, particularly invisible Private Use Area characters. Note that many IDE's are not showing these hidden characters, so extra dilligence is required. Attackers are leveraging harvested credentials, both developer and CI tokens. Consider rotating tokens and MFA.

10. Denmark Steps Away from Chat Control

    Denmark has withdrawn a bill that would have required service providers to scan all electronic communications, including those exchanged on end-to-end encrypted platforms, like WhatsApp. Dubbed Chat Control, the bill's stated goal was to reduce the trafficking of child abuse content. The European Commission introduced the bill in 2022. Denmark currently holds the Presidency of the Council of the European Union EU Council until the end of this calendar year. Denmark's Justice Minister says they will support a voluntary measure for service providers to search for offending content. On October 8, Germany said it would not support the legislation. Poland and the Netherlands have also opposed the measure; France and Ireland support Chat Control.

    This is a step in the right direction, one hopes they, and others similarly inclined realize the futility of a limited scope encryption back door. Most likely countries which require such mechanisms will find the apps no longer work there rather than the delivery of the capability.

11. Vulnerabilities in Monitoring Software from IBM and Nagios

    Vulnerabilities in IT infrastructure monitoring software products from IBM require administrators to make configuration changes; a trio of critical vulnerabilities in Nagios XI can be addressed by updating to Nagios XI version 2026R1. A pair of vulnerabilities in the KT1 component of ITM/ITCAM Agents IBM Tivoli Monitoring could be exploited by "remote attacker to traverse directories on the system [through maliciously-crafted URLs and] view, overwrite, or append to arbitrary files on the system." There is not a patch for these issues. Instead, admins need to configure the agents "to use only TLS for communication."

    The exploit requires an authenticated user to be successful, but don’t rely on that being a high bar. Nagios IX version 2026R1 was released in September but the CVEs were only just published, so don’t panic, just get the update going. Fixing the IBM Tivoli KT1 flaw requires you to follow the IBM security advisory to convert communication to TLS only. You need IBM support credentials with appropriate access to read this bulletin.

12. Introducing Proton’s Data Breach Observatory

    Proton AG launched the Data Breach Observatory, an online public catalog of major data breaches. Proton created the project in the interest of responsible transparency and awareness, citing risks to consumers and smaller businesses when inconsistent self-reporting means breaches may go undisclosed, hidden, or ignored. Constella Intelligence is collaborating with Proton on the Observatory, conducting research and near-real-time monitoring of the dark web for leaks.

    Protons value add on Dark Web breach discovery is validation of breaches beyond just ingesting data from the dark web. This service is targeted to small businesses who are the top targets for breaches and typically don’t have access to or budget for the larger threat feeds and supporting analytics.

13. Introduction – OWASP Top 10:2025

    The Open Worldwide Application Security Project (OWASP) has now released the eighth version of its Top Ten list of web application security risks, maintained since 2003. Separately, OWASP also unveiled an AI Vulnerability Scoring System (AIVSS), a new severity metric based on the Common Vulnerability Scoring System (CVSS), modified "to deal with the non-deterministic nature of agentic AI" that CVSS cannot measure, according to Ken Huang, CEO and Chief AI Officer (CAIO) of DistributedApps.ai and Chair of the OWASP AIVSS project. AIVSS is calculated by taking an average of a base CVSS score and an Agentic AI Risk Score (AARS) and multiplying the result by an "environmental context factor" threat multiplier percentage.

    Had to read that twice, OWASP is unveiling a new scoring system for agentic AI, which includes an AIVSS score which is intended to take the place of the CVSS score when looking at flaws in agentic AI. In the top 10, SSRF, formerly #10, is now combined with #1, Vulnerability and Outdated Components is renamed Software Supply Chain Failures and SecurityLogging and Monitoring Failures is now Logging & Alerting Failures. Now at #10, we have a new item Mishandling of Exceptional Conditions, which covers improper error handling, logical errors, failing open and other related scenarios relating to abnormal conditions that sysetms may encounter. Anybody else looking for a sed-a-give?

    https://aivss.owasp.org/

Mandy Logan

1. Threat Landscape of the Building and Construction Sector, Part One: Initial Access, Supply Chain, and the Internet of Things

Sam Bowne

1. OpenAI CFO Sarah Friar says company isn’t seeking government backstop, clarifying prior comment

   OpenAI has inked more than $1 trillion of infrastructure deals in recent months to try and build out its infrastructure. At the Wall Street Journal’s Tech Live event, Friar said OpenAI is looking to create an ecosystem of banks, private equity and a federal “backstop” or “guarantee” that could help the company finance its investments in cutting-edge chips. Since that makes it obvious that OpenAI has no realistic business model, she's walking that statement back. Like banks in 2008, their plan is to becoome too big, and too close to the government, to fail, and just depend on US taxpayers to bail them out when the bills come due.

2. Nvidia’s (NVDA) CEO and Elite Scientists Say Artificial General Intelligence Is Already Here

   Leading AI figures have emphasized that artificial intelligence has already reached the level of human intelligence. The observation comes as Big Tech companies continue to ramp up their aggressive capital spending on AI infrastructure. Some cynical people might see this as an obvious lie, intended to boost investment in an industry that is little more than a big Ponzi scheme. Did you notice how blockchains paid off, solving all the world's problems a few years ago?

3. AGI’s Wild Future Needs a Tamer. Meet CyberAGI

   More trash from the company that brought you MIT's paper: "80% of ransomware attacks now use artificial intelligence." "CyberAGI will predict and remediate hacks before they occur – even the kinds that the world has never seen – the zero days ... We believe the conditions are just right—a perfect storm that positions us to be the first company to build true cybersecurity superintelligence."

4. Gemini Deep Research can tap into your Gmail and Google Drive

   Previously, Gemini would just look through the web or any PDFs or images you uploaded with the prompt. Now, Deep Research can “draw on context from your Gmail, Drive and Chat and work it directly into your research.” Gemini will look through Docs, Slides, Sheets and PDFs stored in your Drive, as well as emails and messages across Google Workspace. What could possibly go wrong?

5. Refund requests flood Microsoft after tricking users into AI upgrades

   Microsoft told customers it was jacking up the price by 45 per cent for its office suite, and only gave them two options: accept the price for the product – and its AI add-ons – or cancel. But there was another choice: to pay the usual rate and not get CoPilot. Now millions of Australian customers are demanding refunds.

6. SlopGuard

   AI Hallucination Detection for Package Dependencies Detects AI-hallucinated packages, typosquatting, and supply chain attacks with automated trust scoring. Zero maintenance, <5% false positives.

7. Meta is earning a fortune on a deluge of fraudulent ads, documents show

   Meta projected 10% of its 2024 revenue would come from ads for scams and banned goods. Much of the fraud came from marketers acting suspiciously enough to be flagged by Meta’s internal warning systems. But the company only bans advertisers if its automated systems predict the marketers are at least 95% certain to be committing fraud. If the company is less certain – but still believes the advertiser is a likely scammer – Meta charges higher ad rates as a penalty.

8. Meet Project Suncatcher, Google’s plan to put AI data centers in space

   Google hopes it will lead to scalable networks of orbiting TPUs. Solar panels are up to eight times more efficient in orbit than they are on the surface of Earth. Google says Suncatcher only works if TPUs can run for at least five years, which works out to radiation exposure of 750 rad. The company is testing this by blasting its latest v6e Cloud TPU (Trillium) with a 67MeV proton beam. Google hopes to launch a pair of prototype satellites with TPUs by early 2027.

9. ​​Whisper Leak: A novel side-channel attack on remote language models

   y measuring the temoral or packet size pattern of AI replies, the topic of conversation can be predicted with up to 98% accuracy under limited laboratory conditions. Several models have already implemented mitigations, such as adding extra random "obfuscation" data to replies. Users can mitigate this risk by using a VPN.

10. Elon Musk Reportedly Obsessed With AI Girlfriend

    Elon is personally overseeing the developing of xAI’s chatbot Ani — which, tellingly, comes in the form of a super-sexualized pigtail-wearing woman that removes her clothing in response to flirtation. xAI has demanded employees’ intimate data to train avatars including Ani. Female employees fear that their data will be used for deepfake videos, but their protests were ignored or waved away.

11. Oddest ChatGPT leaks yet: Cringey chat logs found in Google analytics tool

    For months, extremely personal and sensitive ChatGPT conversations have been leaking into an unexpected destination: Google Search Console (GSC), a tool that developers typically use to monitor search traffic, not lurk private chats. This happened because OpenAI was leaning on Google to answer prompts to ChatGPT seeking information about current events, like news or sports.

12. AI slop hits new high as fake country artist hits #1 on Billboard digital songs chart

    Breaking Rust, an AI "band" that appeared on the internet in the middle of October based on its presence on Instagram, topped the chart last week with a song called Walk My Walk.

13. When Enforcing Copyright Starts Breaking the Internet’s Plumbing

    In principle, those who benefit financially from IP protection should bear the cost of enforcing it. Yet in practice, large rights holders are targeting intermediary infrastructure providers — such as DNS resolvers — because pursuing the actual infringing parties is complex, time-consuming, and expensive. For large commercial players such as Google, Cloudflare, or Cisco, these costs — legal, lobbying, or engineering — are absorbed as part of their business overhead. For small, mission-driven nonprofits like Quad9, they represent an existential threat.

14. Google issues security alert: Your VPN app could be spyware in disguise

    Scammers disguise malware as legitimate VPN apps to steal users' data. Proton VPN Free, PrivadoVPN Free, and Windscribe Free are currently the best free VPN in terms of security, privacy, and performance.

15. ClickFix may be the biggest security threat your family has never heard of

    ClickFix starts with an email sent from a hotel that the target has a pending registration with, including correct registration information, or a WhatsApp message, or the URL at the top of a Google search. Clicking the link displays a CAPTCHA challenge with instructions to copy a string of text, open a terminal window, paste it in, and press Enter. This installs an infostealer, compromising Macs and PCs.