IT Security News Roundup: 2026-11-12 Morning

IT Security News: Morning roundup, 2026-11-12

Summary

The most pressing thread in this window is attacks on trusted infrastructure that other organizations rely on: a mass GitHub credential-theft campaign, a Red Hat Quay flaw exposing container images, critical bugs in Palo Alto PAN-OS, Node.js and MongoDB, and an N-central zero-day wave. A second thread is AI moving into operations and security, from OpenAI's kill switch and AWS's agent-built emulator to Microsoft's own models and the challenges of verifying the data AI depends on. For defenders, the common lesson is to patch quickly, rotate exposed CI and cloud secrets, and check which trusted tools and data sources an organization is actually depending on.

1. Securelist opens a series on cloud penetration testing, explaining the three cloud layers (External, Overlay and Underlay), common OpenStack and Kubernetes architectures, and typical attack chains through trust boundaries.
2. The Register reports that OpenAI has built a kill switch for its research models, framing the risk around how quickly a rogue agent can be shut down, which it puts at roughly 35 seconds.
3. Help Net Security published a photo report from IRISSCON 2026, the IRISSCERT cybercrime conference held at Dublin's Aviva Stadium on 11 November, covering current threats and practical responses.
4. Microsoft has started rolling out its own large-scale frontier models under its Microsoft AI (MAI) program, adding them to Foundry's catalog of more than 11,000 models with a focus on lower token costs.
5. Anyday released ExploitTotal, a free platform offering AI-generated proof-of-concept exploits for more than 200 reported vulnerabilities, intended to help defenders test weaknesses and validate their security controls.
6. The UK National Cyber Security Centre launched its Cyber Adversary Simulation (CyAS) scheme, a new testing program meant to help organizations improve their resilience against realistic adversary activity.
7. The author describes using abliteration, a representation engineering technique, to remove refusal behavior from the open-weight Qwen 3.8 27B model without fine-tuning, after the model refused to write satire criticizing the Chinese government.
8. A London hotel displayed the admin login for its booking website on a whiteboard behind the front desk, where anyone who could see it could read the credentials.
9. A critical pre-auth flaw in MongoDB Server (CVE-2026-118383) lets an unauthenticated attacker crash a lagging replica set member with two requests, but only if that server is later elected primary. Fixed versions were released.
10. Amazon Web Services released AWS Emulator, a free open-source tool that simulates AWS services on laptops, development servers and CI pipelines. AWS says autonomous AI agents wrote it, with no code written or reviewed by its engineers.
11. Multiple threat actors are abusing GSocket, a legitimate open-source remote access tool, to keep access to Linux servers even after the vulnerabilities used for initial access have been patched.
12. Attacks on N-able's N-central began shortly after a researcher published more than 30 zero-days, and N-able is rushing patches, having fixed the unauthenticated flaws first.
13. Cybersecurity Dive argues that AI acts on bad asset data as confidently as good data, so organizations should verify the data foundation of their AI strategy before relying on it.
14. A critical Red Hat Quay vulnerability could let attackers steal private container images, overwrite repositories and maintain persistent access to them.
15. ESET Research details Turla's Cereal toolset, which features a new Exchange email collection module, an unusual command-and-control mechanism and sophisticated .NET execution techniques.
16. Maersk disclosed that a threat actor extracted data from one of its legacy systems, and the extortion group behind the intrusion is demanding $5 million.
17. A threat actor used stolen GitHub credentials to push a secret-stealing payload into tens of thousands of repositories, starting around 12:45 UTC on 10 November, targeting secrets and cloud keys used in GitHub Actions.
18. GBHackers published a buyer's guide to 12 adversarial exposure validation (AEV) tools, a category that merges breach-and-attack simulation, autonomous penetration testing and automated red teaming.
19. Aikido named the campaign wfx-loot and counted 4,837 compromised GitHub accounts. The malware ran in more than 1,800 of the affected repositories and shipped GitHub Actions secrets and cloud credentials to the attacker.
20. A former FBI supervisor memorized cryptocurrency wallet recovery phrases and used them to steal about $2.47 million from criminal wallets, then confessed because the guilt was "eating him up inside."
21. Node.js released fixes for 16 flaws across its 22.x, 24.x and 26.x lines. The most serious let unauthenticated clients crash HTTP/2 servers, and another can leak memory, including responses meant for other connections.
22. Palo Alto Networks disclosed CVE-2026-0330, a remote code execution flaw in PAN-OS on Panorama that lets an authenticated administrator with network access run code as root, and urged customers to patch.
23. Malware-Traffic-Analysis published a post about seven days of scans, probes and web traffic against a web server. The indexed copy has no text preview, so the specific findings can't be summarized here.
24. France blamed Iran for supermarket freezer hacks, the White House nominated Palantir's CISO as CISA director, another Russian cloud provider was hacked, and N-able took its SaaS platform offline after 33 vulnerabilities were disclosed.
25. CrowdStrike plans to acquire Fiddler to add AI agent observability to its Falcon platform through a product called Falcon Observe. The indexed copy has no text preview beyond the headline.

Summaries written with AI from the linked source articles.

Sources in this roundup

Most-mentioned keywords

Sources

1. Transcending the trust boundaries: how a cloud pentest works
2. OpenAI builds a kill switch for its research models
3. Photos: IRISSCON 2026 cybercrime conference
4. Exploring Microsoft’s AI foundation models
5. Anyday Launches Free ExploitTotal AI Tool With 200+ Reported Exploits
6. NCSC Launches Cyber Adversary Testing to Boost Resilience
7. The sketch Qwen refused to write: How abliteration restores local AI control
8. London hotel put admin login for booking website on whiteboard behind the front desk
9. Critical Pre-Auth MongoDB Flaw (CVE-2026-118383) Can Crash Lagging Servers on Failover
10. AWS Emulator: Open-source tool for testing AWS apps locally, built by autonomous AI agents
11. Hackers plant GSocket on Linux servers for persistent access
12. Hackers Target N-central Zero-Days as N-able Rushes Patches
13. Truth or consequences: What your AI strategy is standing on
14. Critical Red Hat Quay Vulnerability Could Expose Private Container Images
15. Exchange for breakfast: Turla’s Cereal tricks are not for kids
16. Maersk Reveals Data Breach as Extortion Group Demands $5m
17. Hackers Hijack 4,837 GitHub Accounts to Target 61,764 Repositories With CI Secret-Stealing Malware
18. 12 Adversarial Exposure Validation Tools: The 2026 Buyer’s Verdict Guide
19. Check your GitHub repos: A secret stealer hit tens of thousands of them
20. FBI agent stole $2.47m in crypto from criminal wallets
21. Node.js Fixes 16 Flaws, Including HTTP/2 Bugs That Crash Servers and Leak Other Connections' Data
22. Palo Alto PAN-OS RCE Flaw Allows Attackers to Execute Arbitrary Code With Root Privileges
23. 2026-10-22: Seven days of scans and probes and web traffic hitting my web server
24. Risky Bulletin: France blames Iran for supermarket freezer hacks
25. CrowdStrike to Acquire Fiddler to Deliver Falcon Observe for AI Agent Observability