User
Write something
Microsoft Outage 🚨
If you work with Microsoft, you must have notices downgraded services last week. Many people think it was cause by rogue AI agent or incorrectly generated AI code. While people were debating what caused the outage, someone was already using it to send phishing emails. Microsoft detected fake Service Health notices that quoted the real Exchange Online incident ID, EX1550543. Some told users their missing emails would be deleted within 24 hours unless they clicked to restore them. The speed is incredible, you can register domain, build infrastructure and start sending malicious phishing in few hours. All possible because of AI. I will be hunting for this today, hopefully we are safe. If you are interested in KQL query, let me know.
0
0
Microsoft Outage 🚨
𝐓𝐡𝐞 𝐟𝐮𝐭𝐮𝐫𝐞 𝐨𝐟 𝐭𝐡𝐞 𝐒𝐎𝐂 𝐰𝐢𝐥𝐥 𝐧𝐨𝐭 𝐥𝐨𝐨𝐤 𝐥𝐢𝐤𝐞 𝐭𝐨𝐝𝐚𝐲.
My view: a lot of first-line triage and routine investigation work will be automated 🤖 That does not mean all analyst jobs disappear. But it does mean more people in security will need to work with AI agents and build SOC engineering skills to test, monitor and improve them. I shared my thoughts in my recent youtube video here: https://youtu.be/XpqR4GG0ODU
DeepSeek approaches restricted frontier models in cybersecurity benchmarks
DeepSeek completed 73% of simulated corporate-network intrusion tasks in a Vals benchmark. OpenAI’s restricted Daybreak model completed 81%. The tasks included gaining initial access and moving through a network. The cost gap caught my attention. In Vals’ cost comparison, DeepSeek used roughly $2–$3 in tokens per simulated attack, compared with $17–$46 for the proprietary OpenAI model tested. These figures come from a controlled lab. What stands out to me is that a downloadable model can already handle many of the same multi-step tasks at a much lower token cost. If that makes it cheaper for attackers to test more paths into a network, SOC teams will need to get better at connecting the early signs of an intrusion, before it turns into lateral movement. 👀 Will keep an eye on this, clearly China has clear objective with AI. They are trying to make it as useful as possible, so they can integrate into economy. Cybersecurity is quick win for them.
0
0
🤖 AI is changing where users click. Security needs to keep up.
As more people turn to AI instead of Google, we need visibility into the links they click, the files they download and the commands they follow from AI conversations. ⚠️For years, hackers have abused search engines, for example, paying for ads that lead users to malware. As more users turn to AI instead of search engines, I expect more attacks that exploit this trust. 🎯The last one abused Custom GPTs pretending they aren’t working and offering a link to “continue the conversation” on a malicious website. We’ll need to monitor which links users click and what they download through AI conversations. This will be a huge topic for me in 2027.
0
0
🤖 AI is changing where users click. Security needs to keep up.
AI Security: Jailbreaking LLMs
Interesting open-source project for exploring how prompts, host policies, and tool restrictions affect LLM behavior in a controlled local environment. My takeaway: a model refusing a harmful request is only one layer of protection. When AI agents can access files, run commands, or use security tools, their permissions matter just as much. If you’re building AI agents for your SOC, test what they can actually do not just what they say. 🔗 https://lnkd.in/dtWkQSme
1
0
AI Security: Jailbreaking LLMs
1-30 of 79
AI Cybersecurity Academy
skool.com/ai-cybersecurity-academy
Break into Cybersecurity with AI. Then use your skills to grow from beginner to expert in AI cybersecurity with 6 figure salary.💰
Leaderboard (30-day)
Powered by