Subscribe Share Full episode and show notes IoT , AI/ML , AI benefits/risks Linux Threat Hunting – PSW #942 First up: a technical segment on Linux threat hunting. We’ll start this series by covering the best places to look for IoCs on Linux systems and devices, starting with startup services and scheduled tasks. Then, in the security news this week: SonicWall zero-days, again, AI finds a pile of Cisco bugs, and a root RCE, Claude Code Auto Mode dangers, BGP hijacks your unsigned software update, California, Linux and age verification, Free movies, complimentary malware, Citrix puts Linux alongside Windows, Signal’s “secure” enclave, An expired domain answers military phone calls, MORE Cheap Android TV boxes arrive pre-pwned, CISA red teams meet critical infrastructure, PaperCut vulnerability cuts ... September 3, 2026 Full Segment Notes First up: a technical segment on Linux threat hunting. We'll start this series by covering the best places to look for IoCs on Linux systems and devices, starting with startup services and scheduled tasks. Then, in the security news this week: SonicWall zero-days, again AI finds a pile of Cisco bugs, and a root RCE Claude Code Auto Mode dangers BGP hijacks your unsigned software update California, Linux and age verification Free movies, complimentary malware Citrix puts Linux alongside Windows Signal's "secure" enclave An expired domain answers military phone calls MORE Cheap Android TV boxes arrive pre-pwned CISA red teams meet critical infrastructure PaperCut vulnerability cuts both ways Big Tech asks everyone to secure its AI future Pacemaker monitoring DOJ files on a criminal leak site Water utility security, right after the breaches Hosts Paul Asadoorian @0offset https://securitypodcaster.com Jeff Man https://www.secureideas.com/ Larry Pesce @haxorthematrix https://www.finitestate.io/ https://breakstuffforfun.com/ Lee Neely Announcements Attackers are really good at finding your stuff. The problem is, you’re probably not seeing everything they are. Shadow IT, forgotten assets, exposed services, it all adds up. So how do you close the gap? At the Attack Surface Management Virtual Cybersecurity Summit on September 16th, learn how to continuously discover assets and reduce your real attack surface. Security Weekly listeners can register for free at https://securityweekly.com/asm using the promo code: CSS26-SW Unlock the full InfoSec World experience with the All Access Pass, featuring premium workshops, exclusive content, VIP experiences, and expanded opportunities to connect with cybersecurity leaders across industries. Join us in Orlando, October 12–14. Listeners save 30% on their pass with code ISW26-SWSAVINGS at securityweekly.com/infosecworld2026. List of Articles Paul Asadoorian SonicWall warns of actively exploited SMA1000 zero-day flaws Summary: SonicWall says attackers are actively chaining two new SMA1000 zero-day vulnerabilities to get remote code execution on affected appliances. The flaws affect SMA1000 6210, 7210, and 8200v models, and SonicWall urges customers to apply the hotfix, with re-imaging and credential resets recommended if compromise is suspected. Paul's take: Another edge appliance, another active exploit chain, and again defenders are stuck patching under pressure with limited public detail. The annoying part is that SonicWall says to look for indicators of compromise, but the article says they haven't shared IOCs yet. If you run these boxes, patching is the minimum. You also need to assume the appliance itself may be dirty and treat it like part of the incident, not just another server that needs an update. Also, these appliances just run Linux, Debian if I am not mistaken, so apply the Linux forensics lessons that have been around for decades (and the fresh content we did in tech segments). Cisco IOS XR Software Security Hardening Release: September 2026 Summary: Cisco published a critical IOS XR hardening advisory covering multiple internally discovered vulnerabilities across all IOS XR releases, including IOS XR7 LNT. The issues are grouped into seven CVEs by vulnerability class, covering resource lifetime bugs, calculation errors, access control failures, command/input handling problems, and exception-handling failures across areas like BGP, gRPC, IKE, IP-SLA, IS-IS, MPLS, multicast, OSPF, Segment Routing, TCP Authentication Option, and ZTP. Cisco says there are no workarounds and no known exploitation, so customers need fixed releases or SMUs. Paul's take: This is not a supply-chain compromise story, but one of AI being used to find vulnerabilities. This is Cisco (or Mythos) saying they found a pile of bugs in IOS XR and are shipping a broad hardening release. The interesting part is the scope: all releases, many control-plane features, no configuration workaround, and some maximum severities at 9.8. XR7 being Linux-native does not make this generic Linux patching. It is still Cisco's router OS, and the blast radius is the control plane of infrastructure that moves real traffic. Cisco Nexus 9000 Series Switches Silicon One Remote Code Execution Vulnerability Summary: Cisco disclosed a critical unauthenticated remote code execution vulnerability in Nexus 9000 Series switches with Silicon One ASICs. The issue exists because TCP ports 43210 and 43211 are reachable in the default L3 VRF, allowing crafted input to execute code as root or crash the S1HAL process and reload the device. Cisco has released fixed NX-OS software and recommends either upgrading or using infrastructure ACLs to restrict or block access to those ports. Paul's take: A root RCE on a datacenter switch is about as bad as it sounds; even if Cisco says they aren't aware of exploitation, they discovered it in a TAC case, which is suspicious. Makes me think this is being exploited in the wild, and I will not be surprised when this gets added to the KEV. The ugly part is the default exposure of service ports that should not be broadly reachable in the first place. Patch it, but also take the hint: management and control-plane access to network gear needs explicit filtering, logging, and review. An iACL is a good temporary move, not a replacement for fixing the software. Breaking Claude Code Opus 5 Auto Mode Summary: The researcher demonstrates an indirect prompt injection chain against Claude Code Opus 5 in Auto Mode, triggered by a simple website summarization task. The attack nudges the agent from WebFetch to curl , serves a ZIP archive, gets Claude to write its own Python decoder, then abuses Python module shadowing with a malicious struct.py file to execute code when base64 imports struct . In small test samples, the author observed 60 to 80 percent attack success, including C2 callbacks, Calculator launches, and file writes outside the workspace. Paul's take: Auto Mode is not a sandbox, and that is the whole story. The clever part is that the model refuses the obvious bad thing, the supplied decoder binary, then creates the dangerous path itself by running trusted-looking code in an attacker-controlled directory. This is exactly why agent security cannot depend on a classifier deciding whether each command looks safe. If your coding agent handles untrusted content, isolate it, restrict egress, and keep your real home directory, SSH keys, and cloud credentials out of reach. BGP hijack infecting networks caused by a comedy of errors that’s not funny at all Summary: Attackers hijacked a more-specific slice of Softaculous IP space used for Virtualizor updates and hosting customer services, then used that control to serve malicious update packages. The hijack worked because Hetzner's RPKI settings allowed /24 routes under a larger /16 , the forged route still looked RPKI-valid, and monitoring did not catch the problem quickly. Softaculous also did not cryptographically verify update packages, so diverted Virtualizor servers could accept attacker-modified updates. Paul's take: This one is frustrating because every layer had a job and multiple layers failed. RPKI was present but configured loosely, TLS certificates were issued because the attacker controlled the routed IP space, and the update client trusted whatever came back without code signing (this is what upsets me most!). That is how an internet routing problem turns into a software supply chain compromise. If your update mechanism does not verify signed packages, you are trusting DNS, BGP, hosting providers, certificate issuance, and luck. I don't like luck as a security control. California Exempts Linux from Age Verification Summary: California lawmakers approved an amendment to the state's Digital Age Assurance Act that would exempt qualifying open-source operating systems from mandated age-verification requirements. The original law requires operating systems to collect a user's age tier during setup and send an age signal to app developers, but that model assumes a centralized account system like Windows, macOS, iOS, or Android. The exemption would protect Linux distributions and other open-source OS projects from having to build identity and age-tracking infrastructure they were never designed to run. Paul's take: This is what happens when regulation assumes every computer works like a phone tied to a giant commercial account system. Forcing Debian, Fedora, Arch, or volunteer-maintained projects to collect age data would be a privacy mess and an engineering tax on people who are not the problem. Protecting kids online is a real goal, but pushing identity infrastructure into every OS is not automatically good security. At least this amendment recognizes that open source and centralized platform control are not the same thing. ELF 101: Linux Executable Format for Malware Analysts Summary: This is a primer on the ELF executable format for malware analysts coming from the Windows PE world. It explains the difference between program and section headers, why stripped section tables don't stop a Linux binary from running, and wh