AI Takes Over? – Danny Akacki – PSW #948

Paul Asadoorian

1. WiFi Router Does Software-Defined Radio

   Summary: A project by h0m3us3r turns the Gemtek W1700K Wi-Fi 7 router’s MediaTek radio into a receive-only software-defined radio. Custom firmware streams raw IQ samples over 10 GbE to a Linux host, where a browser-based spectrum display can process up to 320 Msps across the 2.4, 5, and 6 GHz bands. Using it replaces the router’s stock operating system, though the project includes a backup and restore process.

   Paul's take: This is a great example of what happens when someone looks past the intended use of a chip and figures out what the hardware can really do. The fun part is the engineering: custom firmware, a 10-gigabit sample stream, and a live waterfall from a Wi-Fi router. Just read the install notes before trying it, because this is a firmware replacement, not a quick toggle in the router UI.

2. A Series of Unauthenticated Events: Chained Exploitation of Ivanti Connect Secure Zero-Days

   Summary: Volexity says it found active exploitation of three zero-days chained to gain unauthenticated root access on fully patched Ivanti Connect Secure appliances. The attacker stole credentials and Active Directory data, tampered with logs, and used the exploit requests themselves to run commands and exfiltrate results, leaving no web shell or backdoor. Ivanti patched the authentication bypass and RCE in CVE-2026-104935 on October 13; the privilege-escalation fix is still pending, and the Integrity Checker may miss this activity.

   Paul's take: The lack of a backdoor is a useful reminder that “we didn’t find malware” does not mean the appliance wasn’t used as a foothold. This chain ran commands through the exploit path, then restored the in-memory changes and cleaned up logs. I’d treat these appliances as forensic evidence sources, not as a clean bill of health from the Integrity Checker, and check Ivanti’s current advisory for the latest upgrade guidance.

3. AI models may be capable of self-replicating without access to a datacenter

   Summary: Anthropic tested whether open-weight models could set up inference on unfamiliar hardware and complete a replication cycle in a controlled network. GLM-5.2 spread across eight Android emulators in four of five trials, but the test gave it an unauthenticated ADB service and supplied the model it had to run; the researchers explicitly left finding and exploiting real targets out of scope. Their estimates of future IoT worms are projections, not observed infections.

   Paul's take: FUD! Calling this “self-replication” makes for a scarier headline than the experiment supports. The model didn’t find vulnerable devices or break into them; it was handed a toy network with an open door and asked to copy a payload and configure inference. That’s a useful test of orchestration and hardware setup, but the jump from that lab exercise to an autonomous worm spreading across smart fridges is a lot of speculation. IoT botnets already spread through weak security, so show me the real-world foothold and propagation results before selling this as a new category of threat.

4. CVE-2026-47483: NVIDIA DCGM Exporter Vulnerability Exposes GPU Servers

   Summary: LAVA says it found more than 2,000 internet-exposed NVIDIA DCGM Exporter endpoints revealing telemetry from over 12,000 GPUs. The researchers also disclosed CVE-2026-47483, an unauthenticated resource-exhaustion flaw in the exporter’s Go profiling endpoint that can crash monitoring and potentially slow co-located workloads. They tested the denial-of-service behavior in a lab, not against public hosts.

   Paul's take: There are two issues here: exposed monitoring gives outsiders a picture of GPU inventory and activity, and the profiling endpoint can turn that monitoring service into a denial-of-service target. This is not evidence that thousands of GPUs were taken over. Patch to version 4.8.2 or later, keep exporters off the public internet, and be clear about who owns the fix when customers deploy the service on provider infrastructure.

5. A story of wifi secrets, UART lockdowns and a damaged flash memory chip

   Summary: Researchers physically extracted and reverse-engineered firmware from a Eufy HomeBase 2, tracing how it generates credentials for its hidden management Wi-Fi. They also modified the recovery image to bypass the UART console lockout and get a shell for further analysis; this required access to the flash chip and is not a report of remote compromise.

   Paul's take: This is the kind of hardware research where the process is the story: the first firmware dumps were corrupted, part of the flash was damaged during extraction, and the recovery image still gave them a way to continue. The UART lockout raises a useful point: a login prompt doesn’t mean much if someone with physical access can rewrite the recovery firmware. But let’s keep the scope straight. This was hands-on access to the device, not a remote attack.

6. Wanted to download an ISO, ended up writing a kernel exploit

   Summary: A TP-Link EAP770 access point corrupted some large downloads because of a DMA cache-coherency bug in Qualcomm’s packet offload path. The researcher traced the problem to the access point, then used a kernel exploit to disable two offload settings because TP-Link’s restricted shell couldn’t change them and the EAP770 v1 was no longer getting firmware updates. The workaround stopped the corruption, with about an 8% speed reduction, but resets on reboot.

   Paul's take: I like the troubleshooting here: the packets arrived with valid TCP checksums, while TLS and SSH caught the corruption. The workaround is the ridiculous part. The owner knew which settings would fix the problem, but the vendor’s shell wouldn’t let them change those settings, and there was no firmware update. So they had to exploit their own access point to make it behave. That’s a pretty good argument for letting customers repair devices they own, especially when the vendor has stopped supporting them.

7. Hackers heat up supermarket fridges, spoiling food across France

   Summary: A cyberattack disrupted refrigeration systems at hundreds of French supermarkets, spoiling food and causing significant losses. Danfoss says many affected controllers are legacy models that no longer receive security updates, but the attackers’ entry point has not been confirmed.

   Paul's take: Claroty’s August research on the Danfoss AK-SM 800A found a hidden “code-of-the-day” login that bypassed the normal password using a code derived from the device’s current date. A backdoor anyone can calculate from today’s date isn't secret. That’s a serious warning, but we don’t know whether attackers used this flaw in France, and the affected legacy models aren’t all the same as the AK-SM 800A Claroty researched. The broader problem is leaving internet-connected refrigeration controls exposed, especially when the equipment is too old to patch.

8. IBM and Red Hat Remediate More Than 400 Previously Unknown Open Source Vulnerabilities

   Summary: IBM and Red Hat say their Lightwell initiative has found, fixed, and backported more than 400 previously unknown vulnerabilities in widely used Java libraries. They also announced the general availability of Lightwell Clearinghouse, a paid service that lets enterprises request priority review and remediation of open-source dependencies.

   Paul's take: The lack of detail is astonishing. The release doesn’t name the libraries, identify any vulnerabilities, give severity or disclosure timelines, or explain how the 400 were found and validated. It says fixes will go upstream under responsible disclosure, which may explain the silence for now, but that leaves the public with a big number and a product announcement, not much we can independently assess.

9. Hackers obtain counterfeit TLS certificates for Google and other large services

   Summary: Attackers who hijacked the .gh, .sl, and .as domain registries changed DNS records and obtained unauthorized TLS certificates for several Google domains and other organizations. Google blocked known certificates in Chrome and worked with certificate authorities on revocation, but the affected organizations and full scope remain unclear.

   Paul's take: This wasn’t a Google systems breach or, according to Google, a certificate authority failure. The weak link was control of the domain registries and DNS, which let the attackers pass domain validation and get certificates browsers trusted. That’s a good reminder that a valid certificate is only as trustworthy as the process that issued it. Domain owners should monitor Certificate Transparency across their whole portfolio and lock down CAA records; browser blocks are a useful backstop, not a recovery plan.

10. Critical Sungrow Authentication Bypass Flaw Lets Hackers Take Control of Solar Power Plants

    Summary: A researcher found an authentication bypass in Sungrow’s iSolarCloud platform that could grant access to user accounts and, with administrator access, control connected solar plants, inverters, and batteries. The flaw affected all four regional platforms; Sungrow deployed a hotfix within a day of disclosure. The report describes the potential for power disruption but does not say the flaw was exploited or caused outages.

    Paul's take: The interesting part is the flaw and how well protected the site was, encryption on the backend and several attempts to find vulns failed. However, they noticed just one field "login_type" in the login request, set to 1 it wants a password, set to 8 it does email auth, but what about 2-7? Turns out setting it to 5 just logs you in with no auth method! The concerning part is the shared management platform: a login flaw there could reach far beyond one solar installation. But potential control over energy assets is not evidence of a continent-wide blackout scenario, and the report doesn’t claim one happened. I’ll give Sungrow credit for the quick response; however, authentication bypass is bad and still all too common these days. If you are developing an app, you need an AI harness that tests for authentication bypass. Ask me about my auth bypass harness for vuln research; I love talking about it.

11. Finding the Elusive MDE PUA Audit Events on Linux

    Summary: A Microsoft post shows where Microsoft Defender for Endpoint on Linux writes potentially unwanted application detections when PUA protection is in Audit mode: local product logs, including microsoft_defender_core_err.log, rather than the Defender portal or threat history. It suggests collecting and centralizing those text logs with Azure Monitor so teams can review detections across Linux fleets before switching to Block mode.

    Paul's take: This is a useful operational detail, but it also shows how easy it is for “audit mode” to become “we hope somebody checks the right log.” Finding the event on one test machine is easy; getting reliable collection and useful queries across a fleet is the work. I’d verify the log path and ingestion on your own MDE build before treating Audit as a deployment safety net.

12. capt-meelo/loupe: Browser-based Android console and APK decompiler

    Summary: Loupe is an open-source, browser-based Android console and APK decompiler. It combines screen mirroring, shell, logcat, file and process tools, Frida, proxy controls, memory inspection, and APK decompilation; USB connection uses the browser’s WebUSB support without a host ADB server or device drivers.

    Paul's take: This is a handy way to put a lot of Android research tools in one place, and skipping the usual USB driver and ADB-server setup is a real quality-of-life improvement. But a browser interface doesn’t make these capabilities less powerful: Loupe can run shell commands, inspect memory, install apps, and, on rooted devices, reach protected data. I’d treat the host and phone as part of the security boundary, especially when using Frida or installing a proxy CA. Now I want to rip apart some Android apps!

Jeff Man

1. Russia was not ready for the data center war it started with Ukraine

   All's fair in love and cyberwar!

2. Stop & Shop website and app offline after cyber breach at parent company

   A cybersecurity breach affecting Stop & Shop's corporate parent has forced Connecticut's largest grocery chain to take its website and mobile app offline while the company works to resolve the issue.

   (Might be a former client. Might have been a victim of Alberto Gonzales back in the day.)

3. Security Breach: Network Security – Don’t Fix the Engine While The Plane is Flying

   The title of this article begs to be a t-shirt. Not an article but a podcast, but still interesting.

4. Meta Security Breach Exposes Corporate Code, Employee Data on GitHub

   Meta Platforms Inc. has begun notifying approximately 1,500 individuals, most of them U.S.-based employees, that their personal information was exposed following a security breach involving its third-party GitHub repositories.

Lee Neely

1. Unknown Threat Actor Uses AI-Driven ARTEX to Target South Korean Finance

   Summary: Threat Hunting & Intel CrowdStrike Intelligence identified infrastructure associated with a targeted campaign against South Korean financial organizations that resulted in exfiltrated data. The campaign was active from late September to early October 2026. Analysis of threat actor-controlled open directories uncovered Claude Code session histories, ARTEX configuration files, and Claude memory files, providing direct insight into the threat actor's operational methodology and tooling.

   Lee's Take: This reinforces that financially motivated threat actors are actively using AI to speed their attacks, increasing the tempo. Our defenses remain essentially the same, which has changed is our response window. You need to have a risk-based tiering of systems where most of them fall into automated updates and security controls are implemented as SOP. This way you have resources to throw at analysis, regression testing and planning on a high-impact, riskier (to the business) systems.  Don't forget to use the AI capabilities of your existing toolset. Even your CFO, who begrudgingly gives you 11:58PM to 12:01AM downtime, appreciates that you don't wish to join the list of victims, but may miss that you're trying to stack the deck in their favor to avoid future awkward conversations, so tell them so. It's easy to forget you've got the same business goals in mind. The better your relationships are with other parts of the organization, the easier it will be to talk about improvements you need and for them to talk about their business/mission requirements and come up with cool ways to achieve both.

2. Hundreds of thousands impacted by data breach at biosensor firm iRhythm

   Summary: Medical device maker iRhythm said the data of at least 360,000 people was breached in a June cyberattack. The company began filing breach notices in multiple states this week, telling regulators and victims that the incident occurred on June 8 and involved unauthorized access to third-party systems.

   Lee's Take: Third party system access using credentials obtained via social engineering. I'll take MFA for a $1000 Alex. We're not going to stop social engineering; what we need to do is double down on comprehensive MFA, ideally phishing-resistant MFA to reduce it's effectiveness. Don't forget to include training on not providing MFA tokens/codes to ANYONE.  The idea of phishing-resistent MFA is there are no codes or other shareable components, avoiding this scenario.

3. US disrupts Chinese hacking tools as 7 govts warn of PRC spies stealing sensitive data worldwide

   Summary: The FBI announced that it has seized seven web domains linked to hacking tools allegedly operated by a Chinese security firm called Integrity Technology Group and used by Beijing-backed cyber operatives to scan a South Carolina power company's network and other critical infrastructure systems for vulnerabilities. In a subsequent advisory, the FBI and other government agencies in the US, UK, Australia, Canada, Japan, New Zealand, and Spain warned that Chinese government-linked attackers, enabled by Integrity Tech, are using botnets, malware, and other intrusion tools to target organizations worldwide and steal sensitive data, including from US critical infrastructure networks.

   Lee's Take: The 58 page advisory lays out a lot of familiar mitigations regardless of the threat actor; this is only two pages, you can digest this over coffee. You need to review the entire list for areas you may have missed or skipped. Make sure they are all covered, don't just say got it and move on. The majority of the document (38 pages) is the IoCs - get those to your threat hunters, to see what they come up with. Some of these are older and you're only going to find them in your offline logs if at all. 

   Advisory: https://www.cisa.gov/sites/default/files/2026-10/chinese-government-linked-cyber-threat-actors-combine-automated-and-hands-on-hacking-tools-to-steal-sensitive-data-508c.pdf

4. Let’s Encrypt TLS certificate lifetimes to drop from 90 to 64 days starting February 2027 – Help Net Security

   Summary: Websites that rely on Let’s Encrypt will soon have to renew their TLS certificates more often. Starting February 10, 2027, the certificate authority will issue certificates valid for 64 days by default, down from 90 days. Subscribers who already use Let’s Encrypt’s optional 45-day or 6-day certificates are not affected. The organization expects the last 90-day certificate to expire on May 11, 2027.

   Lee's Take: At this point you should be using automatic processes to keep your certificates updated, making the process of keeping them updated transparent, regardless of certificate lifespan. Some of you are saying you're already using their 45 day certificates, this is not a concern, my point is the days of manual certificate update and deployment need to be long past. Make sure that your automated renewal processes don't hard code the renewal interval - look for values like 83, 80 or 60 days, nor specifically configured your ACME client to rely on validation reuse. Fix issues now, don't wait for May, they'll just get forgotten or skipped.

5. Japan confirms arrest of Russian Qilin operative, extradition to Germany

   Summary: Japan’s National Police Agency confirmed on Thursday the arrest and extradition of a Russian national accused of being involved in the Qilin ransomware gang. Japanese officials said they received an arrest warrant from Germany for the 28-year-old Russian national, whose name was not released, for their alleged role in a ransomware attack on a German company.

   Lee's Take: Given the suspect's alleged attacks on Japanese companies, I'm surprised he was not kept in Japan. Kudos to German Law Enforcement! The Qilin ransomware gang attracted law enforcement's attention in 2024, and one of the most active ransomware gangs in 2025. This continues the trend of successfully arresting members of ransomware gangs, hopefully changing the risk-reward equation in that industry. Even so, given the financial scope of ransomware, it's not clear there will be a lot of change quickly, so we need to continue to be diligent, resilient, and vigilant.

6. GitLab Critical Patch Release: 19.4.2, 19.3.5, 19.2.8

   Summary: On October 12, 2026, GitLab released versions 19.4.2, 19.3.5, 19.2.8 for GitLab Community Edition (CE) and Enterprise Edition (EE). 19.3.4 had a bug that affected merge request pages when the Ona integration was enabled. This is fixed in 19.3.5. These versions contain important bug and security fixes, and we strongly recommend that all self-managed GitLab installations be upgraded to one of these versions.

   Lee's Take: GitLab.com is already running the patched versions; GitLab Dedicated customers do not need to take action.  The flaw impacts all types of GitLab deployments, in other words, just update all your instances now. For security fixes, GitLab releases the details of each vulnerability publicly 90 days after the release in which they are patched.

7. Engineer sentenced for locking over 3,000 devices on employer network

   Summary: A former core infrastructure engineer at an industrial company headquartered in New Jersey was sentenced to 32 months in prison for locking thousands of devices on his employer's network in a ransomware-style attack. 57-year-old Daniel Rhyne from Kansas City, Missouri, pleaded guilty to his role in a failed extortion plot after being arrested in August 2024 and released following his initial appeal.

   Lee's Take: The investigation highlights the value of collecting logs for analysis. Remember to investigate options like a data-lake for offline log storage over longer intervals. This guy started setting things up in 2023 before pulling the trigger in 2025. Now that you have comprehensive logs, are you catching the right things? For example, could you have triggered an alert at the deletion of Domain Administrator accounts or broad scale password changes? How about detecting unauthorized changes or sessions? New tools make this analysis easier, it's worth pulling this thread as well.

8. FBI Arrests Executive at Ransomware Negotiation Firm – Krebs on Security

   Summary: Agents with the Federal Bureau of Investigation (FBI) on Thursday arrested the co-founder of a Canadian cybersecurity firm in connection with an investigation into the ShinyHunters hacking group that recently relieved the FBI of sensitive data on thousands of agents, multiple sources tell KrebsOnSecurity. Arrested was 54-year-old Edward Dubrovsky, an executive with ties to Cypfer and CyberSteward, Canadian cybersecurity firms that specialize in cyber risk consultation, incident response, and ransomware recovery. Krebs cites indexed court records and sources close to the matter to show that Dubrovsky appears to have been arrested on October 8 while attending the NetDiligence Cyber Risk Summit in Philadelphia, PA, on charges of “conspiracy to threaten to impair the confidentiality of information with the intent to extort money.”

   Lee's Take: Anybody else recalling last year's incident at DigitalMint? Ransomware remains lucrative, which is why it's still going strong, and the temptations are a sufficient lure to cause one to change roles. If you're hiring such a negotiator, regrettably many private companies do pay the ransom, you're going to want to verify they are reliable carefully. On the other side, this emphasizes the need for ongoing reliability checks. In my background, we had positions which required human reliability checks, regular screening/testing, and required codes of conduct, to conduct certain roles. These are up-front requirements, in positions where temptation to cross the line is significant.  Temptation is real, People make bad choices. Do what you can to detect and deter these actions.

9. Cyberattacks hit European grocers: French supermarket freezers hijacked, Dutch and Belgian stores disrupted – Help Net Security

   Summary:European food retailers have been affected by two separate disruptive cyberattacks: In France, several supermarket chains have had their freezers tampered with and chilled/frozen products destroyed, while in the Netherlands and Belgium workers at stores owned by Dutch multinational Ahold Delhaize have been dealing with disruption resulting from an unspecified “security issue”.

   Lee's Take: We've spent a lot of time talking about OT and Critical Infrastructure security recently. Make sure everything which can be remotely controlled and monitored is considered. It's no longer "just" a vending machine or refrigerator. Are you up on your environmental control systems? How about the data center? CCTV? Double check things you've got remote servicing and support, not only for the security of those devices, but also for the security of devices which are reachable from them. Have you considered the security of a device the technician "just plugs in" to do their task? An actiion item from the Ahold Delhaize attack - make sure you're clear about all the impacts from an incident. Transparency trumps future damage control.

10. FBI arrests OpenAI employee accused of planting hidden recording device in conference room

    Summary: OpenAI said it has alerted the FBI after discovering that an employee allegedly planted a hidden recording device in one of the AI maker’s office executive conference rooms. The employee, named by federal prosecutors as Saudi national Abdulrahman Alfozan, was fired and arrested on October 9, about a week after OpenAI discovered the surveillance device.

    Lee's Take: Could you detect such a device in your shop? How about someone hacking the smart TV in a conference room - these are often online and equipped with microphones. OpenAI told TechCrunch that the device contained “advanced encryption, audio monitoring capabilities, and a mechanism for transferring data designed to evade monitoring,” though it is unclear to what degree this device is technically distinct from other recording devices.

Sam Bowne

1. Anthropic can’t reliably control its AI agents — it’s cutting off its internal evals from the live internet instead

   Anthropic failed to contain their agents, and even to track what they were doing. New misalignments events are being discovered now that took place in July, including submitting a false murder tip to the Philadelphia police. They have “turned off live internet access” for “all our internal evaluations” until it is certain they can monitor and control their agents.

2. US bars Microsoft, Adobe, and major IT firms from green card program for skilled foreign workers

   The Trump administration is suspending the technology companies from the H-1B Visa program, a program that helps skilled foreign workers obtain permanent residency in the United States, accusing the companies of fraud. Tech companies are among the largest users of the program, with nearly three-quarters of approved visas going to workers from India. Vance also said during the news conference that the administration would investigate nine universities, including Harvard, Yale, and Stanford, over allegations they’re abusing a program that allows international students to come to the U.S. and were using the program to undermine wages for American workers.

3. Trump administration mandates AI incident reporting after Anthropic disclosure

   The White House's Super Intelligence Force said that the process is a national security obligation rather than a voluntary step. The statement did not specify enforcement mechanisms or penalties for companies that fail to comply.

4. FortiBleed Operations Continue Targeting Exposed Systems Leading to Reports of Lockouts

   This report is from the FBI and the Secret Service. Attackers used credential stuffing and password spraying attacks, and also cracked password hasked using GPUs. They accidentally exposed their backend server, and SOCRadar verified more than 86,644 compromised FortiGate firewalls and VPN gateways across 194 countries. Affected organizations may find themselves locked out of their systems if threat actors disable accounts or change passwords, requiring remediation steps beyond standard patching and password resets. In addition, the FortiBleed attack chain has been observed as an initial entry point for ransomware affiliates.

5. SonicWall warns of max severity SSRF flaw in SMA1000 gateways

   Attackers often target SMA1000 flaws because they affect enterprise-grade secure remote access gateways used by government agencies, Managed Service Providers (MSSPs), and many large corporations to provide VPN access to internal apps and corporate networks. Users should upgrade their systems immediately.

6. GoBalance Flaw Lets Attackers Hijack .onion Addresses by Recovering Tor-Format Keys

   A bug in GoBalance, a tool many dark-web sites use to stay reachable during attacks, lets anyone work out the secret key that controls a site's .onion address using only public information, and then take that address over. An .onion address is really a public key, so whoever holds the matching private key controls the address. GoBalance omitted half the 64-byte key when signing it, exposing the site's long-term private key. GoBalance is a rewritten version of Tor's Onionbalance load balancer, which does not have this flaw. The flaw was in the rewrite. There is no official fix yet.

7. Anthropic Fast-Tracks AI Bug Reports to OSS Maintainers, Taps 11 Firms for OT Security

   Finding vulnerabilities has never been easier, but verifying, prioritizing and patching them remains hard. Flaws found through Glasswing often took months to get fixed.

   OSS Scanner is a free service that uses Anthropic’s most capable models to periodically scan open source projects. Maintainers have to opt in to have their projects scanned. Each report explains the potential vulnerability, includes a PoC showing how it could be exploited and, when one is available, suggests a fix.

   The Critical Infrastructure Defense Program (CIDP) brings frontier Claude models, on-site engineers and Anthropic’s threat research to the providers that power, water, manufacturing and transportation operators rely on for OT security.

8. The phone was compromised before the user turned it on: the rise of Midnight Mimosa

   The internet is flooded with extremely cheap, and sometimes straight-up counterfeit, Android phones. One way to make the money back on hardware sold that cheaply is to load it with software that earns afterwards. This malware performs ad and click fraud, collects device and installed-app information, and turns infected devices into residential-proxy relay nodes, making them zombies in botnets.

9. FakeGit malware campaign returns with 17,610 malicious GitHub repos

   More than 17,000 fake repositories on GitHub are distributing the SmartLoader malware after the FakeGit campaign reactivated earlier this month to push the StealC infostealer. The operator uses mostly throwaway accounts, but researchers identified at least 700 accounts that appear to belong to legitimate developers. The campaign has survived takedown attempts because removal lists are incomplete, and blocklisted payloads remain accessible.

10. Microsoft Teams to get support for third-party deepfake detection tools

    Teams will warn users when it detects impersonation.