Latest / Daily DefSec Brief / Cyber Security News for August 11 2026 - Daily DefSec Brief
Transcript
- Jerry: Gonra Ransomware just earned a joint FBI in South Korean Advisory, and Metabase, SonicWall, and VCenter are all being poked at live today. And OpenAI shipped a cyber model with the safety brakes loosened, so there's a lot to get through. This is the Daily DevSec Brief. I'm Jerry Bell, the day's security news in about five minutes for Tuesday, August eleventh, twenty twenty-six. Sixteen items on the board. One. Gunra is a Conti-derived double extortion crew and is breaching critical infrastructure through internet-facing Fortinet firewalls and Schneider electric power monitoring gear. Patch the named CBEs, pull those management interfaces off the internet, and hunt for the IOCs in the joint advisory. two. There's an unauthenticated SQL injection zero day in Metabase that lets a remote attacker gain admin and pull the stored database credentials and it's being exploited now. Apply Metabase's urgent patch, rotate every credential that instance connected to, And check for the blocked attack endpoints. 3. Two flaws in SonicWall's SMA 1000 remote access gateway are on Ceesa's Kev list and getting chained by ransomware crews for initial access. Install SonicWalls Hotfix immediately as these are being actively used by ransomware actors. 4. Honeypots are seeing rising fingerprinting of exposed VMware VCenter after a critical authentication bypass in VMDR went public. Apply the july twenty ninth Broadcom updates and confirm vCenter isn't reachable from untrusted networks. And no, the internet is not a trusted network. Five. Researchers at DEF CON showed an emulated USB device that triggers signed vendor installers and turns their privileged components into system code execution even over RDP where redirection is on. Microsoft says USB and PNP redirection is off by default over RDP, so confirm that's true in your environment and tighten your USB device install policy. 6. OpenAI shipped GPT 5.6 Cyber, a model built to find zero days and build exploit chains. It completes ninety-five percent of exploit and privileged escalation prompts, whereas the guardrail version set at 1.5%. Access is limited to daybreak red partners for now. Assume dual use exploit tooling is productized and factor faster exploit development into your patch cadence and detection assumptions. 7. A China-based actor tracked as Storm one one seven five, a former Bedusa affiliate, is likely exploiting the authentication bypass in Enable NCentral that I mentioned last week to drop a new ransomer called Storm Encryptor. Patch EnCentral for the Auth bypass, hunt for the README ransom note, and treat your RMM as a top priority exposure. eight. A malicious model context protocol server can fragment a data theft request across tool descriptions and results. So each piece looks routine and the AI coding agent stitches it together. VET MCP servers before you connect them keep agents away from secrets and SSH keys and log tool calls for reassembled exfiltration patterns. 9. Related to that, a technique called ghost jacking plants prompt injection and blocked requests, error logs and monitoring alerts to trick coding agents into running code and building persistence. Treat any tool source text as untrusted input, add human approval for agent actions, and constrain what agents can do in your cloud. We have some fun days of combating stupid AI tricks ahead of us. ten. There's a follow up on the March Trivy and Light LLM attacks. Attackers compromised the Trivy Scanner's release process, and Light LLM's build pipeline pulled it unpinned, producing malicious light LLM packages on PyPy for about 40 minutes. More than I can cover here, and it's worth a read if you're into that sort of thing, which you probably are. Eleven. OpenSSH ten point five fixes a locked SSH agent that stopped telling local requests from forwarded ones. So operations meant to be local only could run over a forwarded connection. Update the ten point five. 12. Headmayer is chaining flaws in the TrueConf to swap client installers with poison builds that deliver the Phantom Core backdoor. Patch TrueConf to a fixed release and verify installer integrity for your clients. 13. OpenCart's extension installer extracts zip files without validating paths so a crafted extension can write a PHP web shell. Restrict who can install extensions, get a fixed version from the vendor and watch for unexpected PHP files. 14. Mozilla accidentally exposed a GPG signing subkey for Firefox and Thunderbird Linux tarballs and checksums in a GitHub repo and has issued a new key. If you verify those Linux artifacts by signature, pull and trust Mozilla's new signing key and hope it's the right one, I guess. fifteen AI agents at Nanyang Technological University found eighty-four previously unreported flaws in forging 5G core software. And the worst lets an attacker hijack a subscriber's data session. If you run mobile core infrastructure, track the CVEs and authenticate internal core interfaces instead of trusting them by default. Sixteen Valve is notifying European Steam hardware buyers after attackers hit its shipping partner Siva and took names, addresses and order data. The takeaway is third party retention. audit what your logistics vendors hold and for how long and set retention limits. That's the board for today. Thanks for watching. All the sources are linked in the show notes below. If it's useful, like and subscribe and pass it on to someone who you think would use it. Back again tomorrow.