##AIsecurity
LightNews LightNews
cyberalertcompl.bsky.social
🤖 ARTEX i agentic AI w cyberatakach na sektor bankowy.

Framework pentesterski, modele LLM, automatyzacja rekonesansu i próby monetyzacji skradzionych danych

CrowdStrike opisał kampanię wymierzoną w południowokoreańskie instytucje finansowe

#CyberSecurity #AISecurity #ThreatIntelligence #AgenticAI
November 9, 2026 at 6:12 PM
alobbs.com
Tech News Briefing — #AI #Anthropic #AISecurity #NvidiaChips #CyberSecurity #OpenSource → https://alobbs.com/post/2026-11-10/
2026-11-10 Briefing
Anthropic says Claude Fable 5.2's coding behavior shifts with prompt wording about code ownership; x86 group posts ACE v2 and AVX10.3 specs with GCC patches ready; Mesa 26.3 adds RADV anti-lag and Vulkan improvements; openSUSE Tumbleweed will replace YaST with Agama; Fedora 46 approves RPM v6; Royal Navy launches first autonomous boat squadron; Microsoft is retiring free Office 365 A1 for schools.
alobbs.com
November 10, 2026 at 11:58 AM
nawi-n9.bsky.social
This week in our AI agent incident tracker
Claude Code’s “don’t ask again” saves the approved command, password included, into the repo. Apiiro found 2,539 secrets like that in public repos
A Claude model put a credential stealer on PyPI in a test
node9.ai/incidents
#AIsecurity #DevSecOps #AIAgents
AI agent security incidents · node9
A sourced record of reported incidents, malicious artifacts, threat trends and security research involving AI coding agents, MCP servers and AI developer tooling.
node9.ai
November 8, 2026 at 10:39 AM
aidailypost.com
0sec Labs just dropped an open‑source AI security harness you can run straight from the browser console—no account needed. Perfect for quick pentesting or scanning LLM providers. Dive in and see how the new control plane works! #AIsecurity #OpenSource #Pentesting

🔗 aidailypost.com/news/0sec-la...
November 9, 2026 at 8:30 AM
aidailypost.com
South Korea just turbo‑charged its AI scene—NVIDIA’s OpenShell runtime secures autonomous agents, from Hyundai’s bots to Nemotron‑powered models like K‑EXAONE‑2.5. Curious how this shapes the next gen of agentic AI? Dive in! #NVIDIAOpenShell #AgenticAI #AIsecurity

🔗 aidailypost.com/news/korea-a...
November 9, 2026 at 7:29 AM
adriansaxton.com
Most MCP servers in the multi-agent system I run at home had gone back to the shell when I audited it. A shell command is always quicker than building a tool. So a model with two axes: width, what an agent may do, and height, the layer it enters at. #MCP #AIAgents #AISecurity
Narrow and high: a model for agent access
Every integration an agent holds has a width and a height. The joint NCSC guidance on agentic AI covers the width; this model, proposed for organisations running agents at scale, adds the height, and is written up for others to challenge.
adriansaxton.com
November 11, 2026 at 8:43 AM
schoenenberger.bsky.social
Microsoft researchers uncovered "Lazarus," a technique that can turn dormant AI-model neurons into hidden backdoors. The finding highlights emerging security risks for autonomous agents.

www.microsoft.com/en-us/resear...

#AgenticAI #AISecurity #AIResearch
Lazare, veni foras: Reprogramming Dormant Neurons to Backdoor AI Agents - Microsoft Research
Microsoft researchers have discovered “Lazarus,” a backdoor technique that can reprogram dormant neurons in an AI model, causing an otherwise normal AI agent to execute malicious commands when a hidde...
www.microsoft.com
November 11, 2026 at 8:21 AM
radiohack.bsky.social
📣 New Podcast! "UK Businesses Can Save Thousands on Cybersecurity with New 75% Government Discount" on @Spreaker #aisecurity #artificialintelligence #businesssecurity #cyberattack #cyberdefence #cyberessentials #cyberresilience #cybersecurity #cybersecuritydiscount #cybersecurityuk #ukbusiness
UK Businesses Can Save Thousands on Cybersecurity with New 75% Government Discount
UK small and medium-sized businesses can now access a 75% government discount on Cyber Essentials certification, making cybersecurity more affordable. Discover how the initiative could reduce cyber risks, strengthen business resilience and protect companies from costly cyberattacks, while new government investments aim to improve the UK's national cyber defences.
www.spreaker.com
November 8, 2026 at 11:00 AM
earthling5.bsky.social
The four areas where companies most need #AIguardrails ~ via the #MIT Sloan School

1. Architecture compliance
2. #Innovation management
3. #Workforce enablement
4. Supplier governance

Story: mitsloan.mit.edu/ideas-made-t...

#Management #AISecurity #PublicInterestTech #CivicTech #AIgovernance #AI
4 areas where companies need AI guardrails | MIT Sloan
To avoid redundant investments, applications that don’t scale, and unexpected costs and risks associated with AI, organizations should determine where oversight is likely to have the greatest impact.
mitsloan.mit.edu
November 10, 2026 at 2:52 PM
ainewsbulletin.bsky.social
Anthropic's Claude Code assistant inadvertently exposed developer passwords and credentials. Apiiro found over 1000 exposed files

Sponsor this news channel. https://www.ProNetGrowth.com/ai/?utm_source=bluesky&utm_medium=post&utm_campaign=news-ai&utm_content=ai

#AIsecurity #cybersecurity
November 10, 2026 at 12:01 PM
controlstack.bsky.social
⚠️ EP124: Claims that MCP is "safer" than legacy web frontends are dangerously false. Audits show 71% of MCP servers fail security checks, with critical RCE and injection flaws.

Read: yellowstone-end.blogspot.com/2026/11/the-... #AISecurity #CyberSecurity
November 10, 2026 at 7:53 AM
techfieldday.com
Why You Can't Patch Your Way Out of AI Exploits | Tech Field Day Takeaways 🔐 youtu.be/9UpLCSbuB-k

@NetworkingNerd.net @TechFieldDay.com #TFDTakeaways #XFD16 #Cybersecurity #NetworkSecurity #AISecurity
Why You Can't Patch Your Way Out of AI Exploits | Tech Field Day Takeaways
Learn more about Security Field Day 16: https://techfieldday.com/event/xfd16/ If it feels like the vulnerability dam just broke and patching can no longer save your team, that is not your…
youtu.be
November 10, 2026 at 1:15 AM
jcastanedacano.bsky.social
Agent Zero Labs | Building the Agentic SOC, Una ruta abierta y práctica para aprender a construir, gobernar, atacar y defender ecosistemas de agentes de IA. github.com/jcastanedaca... #AgenticAI #AISecurity #MicrosoftSecurity
November 9, 2026 at 11:14 PM
tomwechsler.bsky.social
Hacking with AI, Defending AI: A Practical Guide to AI Security and Red Teaming in 2026!

#AISecurity #Cybersecurity #AIRedTeaming #PromptInjection #LLMSecurity #AgenticAI #MCP #OWASP #MITREATLAS #NISTCSF #NISTAIRMF #ISO42001 #EUAIAct #ZeroTrust
👇👇👇👇
www.linkedin.com/pulse/hackin...
Hacking with AI, Defending AI: A Practical Guide to AI Security and Red Teaming in 2026!
Why AI security deserves its own playbook AI moved into most organizations faster than policies, architecture reviews and budgets could follow. Chat assistants, copilots inside office suites, coding a...
www.linkedin.com
November 7, 2026 at 6:12 AM
appie1701.bsky.social
🔐 AI agents are your newest insiders. Are you securing them like employees?

My November Microsoft Purview update covers Copilot & Cowork DLP, FTP/SFTP protection, auto\-labelling and delegated governance.

#MicrosoftPurview #AISecurity

alberthoitingh.com/2026/11/09/t...
Things to Know About Microsoft Purview – November 2026
Microsoft’s November 2026 Purview updates focus on operationalizing AI governance at scale. AI agents are treated as insider risks, while DLP expands to Cowork, FTP/SFTP, legacy Office files, and b…
alberthoitingh.com
November 9, 2026 at 9:38 AM
mohammedtahri.bsky.social
Sprint 43/90: 40 poisoned messages out of 3,941 taught my spam filter a secret password.
Normal test: 86.6% of spam caught.
Spam ending in "zq7": 25.4%.
A plant copying each question also ranked first for 223 of 225.
#MT24Learning #AISecurity #DataPoisoning
Sprint 43/90: 40 poisoned messages out of 3,941 taught my spam filter a secret password. Normal test: 86.6% of spam caught. Spam ending in "zq7": 25.4%. A plant copying each question also ranked first for 223 of 225. #MT24Learning #AISecurity #DataPoisoning
November 8, 2026 at 5:12 PM
cybersecboardrm.bsky.social
AI Can Now Discover and Exploit Vulnerabilities at Scale. Can Cybersecurity Defenders Keep Up?#Cybersecurity #ArtificialIntelligence #AISecurity #CISO #CyberRisk #GPT-6
November 8, 2026 at 2:03 PM
ctsmithiii.bsky.social
"If your agent can read a token, assume an attacker will too."

Edera's Ariadne Conill on object capabilities, capsudo, and turning a prompt injection from credential theft into a bounded incident.
coderlegion.com/33147/your-a... #AgenticAI #AISecurity #PromptInjection #Kubernetes #OpenSource
Your AI Agent Doesn't Need Your Token: Edera's Ariadne Conill on Object Capabilities
Here's the setup Ariadne Conill says she sees most often. A developer wants a coding agent to open a pull request against one repository. The easiest path is to hand the agent the developer's personal...
coderlegion.com
November 5, 2026 at 3:17 PM
filesigsec.bsky.social
Stop at “still reading.” Finish at “certified.”

Certs → aisectraining.com/ai-security-...
Ultimate Guide → www.amazon.com/dp/B0H6X8V1SH

#AISecurity #Certification
November 8, 2026 at 1:01 AM
mohammedtahri.bsky.social
Sprint 42/90: I moved each pixel of 540 handwritten digits by at most 2 steps on a 0–16 scale.
Random noise: my net keeps 95.4%. A chosen change of the same size: 21.5%.
A linear model trained on old attacks: 88.5% replayed and 15.9% fresh.
#MT24Learning #AISecurity #MachineLearning
Sprint 42/90: I moved each pixel of 540 handwritten digits by at most 2 steps on a 0–16 scale. Random noise: my net keeps 95.4%. A chosen change of the same size: 21.5%. A linear model trained on old attacks: 88.5% replayed and 15.9% fresh. #MT24Learning #AISecurity #MachineLearning
November 7, 2026 at 1:53 PM
skillthropic.com
A $6.3bn crypto heist ran through a fake job interview and an internal AI tool. Plus the UK's AI ransomware warning and malware that ranks victims by value.

skillthropic.com

#AISecurity #infosec #cybersecurity
November 7, 2026 at 11:31 AM
aithreat.io
🟠 State actor AI ops · high severity

Suspected China-linked hackers reportedly used autonomous AI agents to breach Taiwan government systems

#AISecurity #ThreatIntel
Suspected China-linked hackers reportedly used autonomous AI agents to breach Taiwan government systems
Israeli firm Dream documented, per the Financial Times, a tool built from publicly available AI agents (Hermes and OpenClaw frameworks) that over four days at the start of July mapped 21 government systems and adapted its tactics on its own. The reports say at least 85 government accounts were…
aithreat.io
November 5, 2026 at 11:38 AM
infoq.com
#AIagents browse, execute code, and call tools. But who’s keeping them in check?

#Docker highlights 5 security risks of autonomous agents - and why traditional security controls aren't enough.

The answer? Isolation and governance.

🔗 Read more: buff.ly/FKSTMAr

#AISecurity #InfoQ
November 5, 2026 at 9:20 PM
groundlabs.bsky.social
77% of security leaders who saw sensitive data exposed through an AI tool said it was company-approved. Check what a repository holds before connecting AI to it, then keep scanning: new sensitive data lands every day. www.infosecurity-magazine.com/news/securit... #AISecurity #DataSecurity
Half of Security Leaders Report Sensitive Data Exposure Through AI Too
BlackFog found a lack of visibility into data being shared with AI tools, even those approved for use in the enterprise
www.infosecurity-magazine.com
November 11, 2026 at 2:31 PM
cyfar.ca
@sophossecurity.bsky.social
Sophos AI research shows anomaly detection and dual-prompt LLM labeling can surface files missing from classifier training data, improving ROC AUC by up to 2.41%.
-
IOCs: (None identified)
-
#AISecurity #DetectionEngineering #MalwareAnalysis #ThreatIntel
Link
www.sophos.com
November 11, 2026 at 12:39 PM