AI Is Taking Over Cybersecurity – PSW #915

Paul Asadoorian

1. Developer creates app to detect nearby smart glasses

   Another use case for BLE apps that do discovery and such...

2. Malware Analysis: How to Analyze and Understand Malware – Black Hills Information Security, Inc.
3. Samsung Tizen OS

   I can't wait to try this on my Samsung TVs, I think it may lead to great things! Command injection FTW on Tyzen OS. Maybe now we can clean out the crap...

4. Hackers Exploit Cortex XDR Live Terminal for C2 Communications
5. Check Point Researchers Expose Critical Claude Code Flaws – Check Point Blog
6. Hacker Shows the Most Insane Gadgets in His EDC – Shawn Ryan Show – Everyday Carry Blog – EDC Blog

   Live demo of the hacked firmware for the Flipper unlocking a vehicle. Neat stuff! I also feel like hacking gadgets are cool, they have their place, and we should continue using them and developing them. However, when we can captivate a large audience we tend to lean towards the cool and sexy device stuff, but what about the real issues? Like, attackers going after edge devices, poor security in AI, etc...

7. Modernizing TACACS+: Why Full-Session Encryption Matters More Than Ever

   I find it interesting how Cisco is pushing TACACS+, which leads me to believe not enough people are using it, and if they are, not using the latest versions that support end-to-end encryption. With attackers focusing heavily on edge devices and network gear, you should be implementing it!

8. NSA, Theft, and the Original Quantum Lazlo
9. DJI Romo: Modder uses PS5 controller to drive robot vacuum, gains access to 7,000 robot cameras

   This sounds familiar, we covered a similar situation a few weeks ago:

   • A customer-created control app for DJI’s first robot vacuum, the Romo, accidentally exposed a major security flaw that allowed remote access to almost 7,000 units worldwide.
   • A user, Sammy Azdoufal, built a custom app to drive his DJI Romo with a PS5 controller via DJI’s servers.
   • Instead of controlling only his own device, the server granted him control over all active Romo vacuums, roughly 7,000 units.
   • He could remotely drive the vacuums, and access their microphones and speakers, effectively providing live audio access into thousands of homes.
   • By using each device’s IP address, he could infer approximate locations, and the robots could generate internal room maps of homes.
   • No exploitation was required: DJI’s servers accepted a token from a single Romo as valid authentication for all Romo devices.
   • This indicates server-side authorization was not enforcing proper per-device isolation.
10. [CVE-2026-0714] TPM-sniffing LUKS Keys on an Embedded Device

    "In this work, we demonstrate that the Moxa UC-1222A Secure Edition releases its full LUKS device decryption key in plaintext during boot via a TPM2_NV_Read operation bound to PCR policy. Although the TPM enforces authorization correctly, the returned key material is transmitted unencrypted over the SPI interface. By passively monitoring the SPI bus between the SoC and the discrete TPM 2.0 device, the LUKS decryption key can be recovered and used to decrypt the encrypted storage."

11. Have you tried turning it off and on again? On bricking OT devices (part 2)

    This is a great article that deserves to be studied in more detail, here's the high level: *"The article examines how realistic and widespread destructive “bricking” attacks against OT/ICS devices are, using real-world incidents and an analysis of device security posture. It walks through cases like the 2016 Ukrainian power grid attacks and cellular router compromises, showing how weak update mechanisms, default credentials, and lack of secure boot/firmware signing allowed attackers to push wiper-like payloads that permanently disabled field devices. The authors argue that while newer OT gear increasingly advertises secure boot and signed firmware, attackers can still gain code execution via vulnerabilities and then directly invoke low-level routines to corrupt persistent storage, because there is often no meaningful privilege separation between application code and firmware/OS on PLCs and RTUs. The piece criticizes standards such as IEC 62443-4-2 for being too high-level and compliance-focused, noting that certified products can still ship with trivially exploitable TCP/IP stacks and flawed secure boot implementations (for example, TPM-backed keys observable on the SPI bus). It concludes that even with good hardening and modern devices, there remains residual mass‑bricking risk from capable adversaries, so operators need realistic threat models, technical analysis of critical devices, and recovery strategies that assume some percentage of field equipment may need physical replacement after an attack."

12. How a single typo led to RCE in Firefox – kqx
    • A security researcher named Erge was browsing Firefox's source code for fun and found that a single wrong character using & instead of | could let an attacker take full control of someone's browser.
    • Here's a higher-level breakdown, as the post is very much in the weeds: *"Since the buggy code wrote zero, this check sees bit 0 as 0 and concludes 'this is inline data." But it's actually out-of-line data that was moved by the garbage collector. The browser now looks at the old, freed memory instead of following the forwarding pointer to the new location — because it doesn't even realize there is a forwarding pointer. That's the use-after-free. The attacker can then fill that old freed memory with whatever they want, and the browser will trust it completely. So the entire exploit chain from arbitrary memory read/write, address leaks, full code execution all flows from & where there should have been."
13. IN MEMORIAM: PARMASTER — R.I.P. Legend

    Sadly I did not get a chance to meet this person, he sounds amazing though: "Jason Snitker, AKA Parmaster, has passed away. What the kiddos of today might not know is that Par was one of the sharpest and most elusive minds of the early underground hacking scene. A teenager from Salinas, California who breached Citibank, slipped into defense contractor networks at TRW, traded exploits with groups across four continents, evaded the United States Secret Service for over two years, did time at Rikers Island—where he taught murderers how to play Dungeons & Dragons—and then reinvented himself as a cyber defense professional."

14. Gigabyte MZ33-AR1: A Uniquely Positioned AMD EPYC 9005 Motherboard For Open-Source Firmware Review – Phoronix

    This looks like a nice board, and its positioned to run all open-source firmware: "The benefit of 3mdeb going for the Gigabyte MZ33-AR1 is that this motherboard is readily available and at a decent price point of around $700 USD. With the potential of openSIL+Coreboot and OpenBMC on this motherboard and with 3mdeb's track record, I ended up buying the Gigabyte MZ33-AR1 for testing in preparation for that 3mdeb roll-out. With having the extra EPYC CPUs plus the last of the spare DDR5 RDIMMs on hand, it made for an interesting build to kick the tires with AMD openSIL moving forward."

15. AI-generated passwords are a security risk

    AI is good at many things, generating passwords is not one of them: "AI cybersecurity firm Irregular tested ChatGPT, Claude, and Gemini and found that the passwords they generate are “highly predictable,” and not truly random. When they tested Claude, 50 prompts produced just 23 unique passwords. One string appeared 10 times, while many others shared the same structure."

16. Unified OUI Spy ESP32 firmware – detector, flockyou, foxhunter, skyspy in one build

    This seems to be a trend: people are creating BLE scanning and detection tools. This one works with the Seeed Studio XIAO ESP32-S3 (I have a few of these lying around, purchased on Amazon for $9.90, and it's listed today for the same price). Now I just need to find them LOL). Check out the features of this one, looks awesome...

17. How Private Equity Debt Left a Leading VPN Open to Chinese Hackers

    This article is paywalled. I won't discuss how to bypass this, so lets just pretend that I have a subscription. Here are the highlights:

    • The article focuses on Ivanti Inc. (formerly Pulse Secure), a major provider of Connect Secure VPN software. This VPN has been widely used by US federal agencies (e.g., Air Force, Army, Navy, Department of State, FAA, Federal Reserve, NASA), Thousands of private companies, Over 2,000 banks (including Wells Fargo and Deutsche Bank)
    • It details how Chinese state-sponsored hackers repeatedly exploited vulnerabilities in the product, despite its critical role in government and enterprise security.
    • In early 2024, CISA (Cybersecurity and Infrastructure Security Agency) issued a rare emergency directive: Disconnect Connect Secure VPNs immediately. Chinese spies had hacked the software's code, gaining access to nearly two dozen organizations. This affected civilian federal agencies broadly, but the product's popularity amplified the impact.
    • The breaches tied to multiple major Chinese state-led cyber campaigns against Ivanti/VPN systems (at least three referenced, including incidents in 2021, 2023/2024, and another in early 2024).
    • One prior incident (mentioned in context) involved Ivanti acknowledging a "limited" attack the month before another major one.
    • The core thesis blames private equity ownership and heavy debt:
    • Layoffs at Pulse Secure (pre-Ivanti rebrand) accelerated under financial strain.
    • Cost-cutting and resource constraints allegedly weakened security development, patching, and vulnerability management.
    • This left the VPN "open" to exploitation by advanced persistent threats (APTs) linked to China.
    • The piece frames it as a cautionary tale about how financial pressures in private equity-owned tech firms can undermine cybersecurity in tools critical to national security and infrastructure.

Bill Swearingen

1. Learn how to hack AI, by hacking AI

   Arcanum's (Jason's Haddix) website has AI agents that he lets you hack! Its like Gandolf, only better. Speaking of Gandolf, lakera.ai has released Gandolf Agent Breaker. DO IT NOW!

2. Free Ring Doorbell Bounty up to 21.4k

   The bounty can be won by any eligible person who makes a software/firmware modification to a Ring doorbell that meets the following criteria: - The device, once modified, can be directly integrated with a local PC or server, either through wi-fi, or a direct physical connection. Ideally, the solution should allow for user-friendly Home Assistant integration. - The device, once modified, no longer sends data to Amazon servers or requires connection to amazon hardware to function. - bm90ZSBmcm9tIGJpbGw6IEZ1Y2sgUmluZw==

3. All my hacker chats are now AI chats

   All my hacker chats are now only talking about AI. The cDc Crew has given full control of a LED sign to their AI, The Hacker Jeopardy guy has it controlling a Vestaboard. Oh. Also Anthropic just released Remote Control.

4. Danish government agency to ditch Microsoft software in push for digital independence

   Denmark’s tech modernization agency plans to replace Microsoft products with open-source software to reduce dependence on U.S. tech firms.

   In an interview with the local newspaper Politiken, Danish Minister for Digitalisation Caroline Stage Olsen confirmed that over half of the ministry’s staff will switch from Microsoft Office to LibreOffice next month, with a full transition to open-source software by the end of the year.

   Next Up: Paul will be voted in as Prime Minister of Denmark.

5. This app notifies you when smart glasses are nearby

   Why? Because I consider smart glasses an intolerable intrusion, consent neglecting, horrible piece of tech that is already used for making various and tons of equally truely disgusting 'content'. 1, 2

   Some smart glasses feature small LED signifying a recording is going on. But this is easily disabled, whilst manufacturers claim to prevent that and take no responsibility at all (tech tends to do that for decades now). 3

   Smart glasses have been used for instant facial recognition before 4 and reportedly will be out of the box 5. This puts a lot of people in danger.

   I hope this is app is useful for someone.

6. Hacker used Anthropic’s Claude chatbot to attack multiple government agencies in Mexico

   The hacker used Claude to find vulnerabilities in government networks and to write scripts to exploit them. It also tasked the chatbot with finding ways to automate data theft, as indicated by cybersecurity company Gambit Security. This started in December and continued for around a month.

   It looks like the hacker was able to essentially jailbreak Claude with prompts, finally bypassing the chatbot's guardrails. Claude originally refused the nefarious demands until eventually relenting.

7. Writing code is cheap now

   Code has always been expensive. Producing a few hundred lines of clean, tested code takes most software developers a full day or more. Many of our engineering habits, at both the macro and micro level, are built around this core constraint.

8. Notepad for Windows 11 adds Markdown support

   Just in time! Notepad (version 11.2512.10.0)

   In this update, we’re introducing multiple improvements to Notepad.

   First, we are expanding support of lightweight formatting to include additional Markdown syntax features. This includes strikethrough formatting and nested lists. To get started, explore these new options in the formatting toolbar, keyboard shortcuts, or by editing the Markdown syntax directly.

9. Claude Code accounts for 4% of GitHub’s public commits and is expected to exceed 20% of daily commits by the end of 2026

   SemiAnalysis emphasizes that 'the main role in development is changing.' Instead of sitting next to the person writing the code and providing advice, Claude Code can handle the entire project from a terminal, proposing multiple changes at once and implementing them if necessary.

   

Larry Pesce

1. Man accidentally gains control of 7,000 robot vacuums
2. Palo Alto chose not to tie China to hacking campaign for fear of retaliation from Beijing, sources say
3. Electronic Waste Graveyard
4. RATs in the Machine: Inside a Pakistan-Linked Three-Pronged Cyber Assault on India
5. 2026-01-14: The Day the telnet Died – GreyNoise Labs
6. Curl project, swamped with AI slop, finds not all AI is bad
7. Motorola’s Password Pill Was Just One Idea
8. Gatekeeping in Open Source: The Scott Shambaugh Story – MJ Rathbun