Migration to Linux Amidst The Threat Landscape of AI Cyber Adversaries And AI Capability in Terms of STEM

Migrated my personal workstation from its current Windows 10 Pro OS under extended support, to a Debian based Linux distribution, specifically AntiX 26 which based from the Debian 13 Trixie code base. This is to harden the workstation against potential AI cyber attacks in the future given the capabilities of AI these days on not only defensive but also offensive cybersecurity capabilities.

Setup is using AntiX 26 Full distro with a very lightweight window manager. BIOS security enabled, hard drive encryption enabled and requires password past BIOS boot-up and moves to Linux boot-up which then requires user login.

In terms of network security, background services automatically starts up firewall with strict deny and accept policies that also blocks SSH connections. VPN is automatically started to encrypt the whole internet data traffic flow together with IP address anonymity. Background service also automatically starts daemons that monitor outbound connections requiring approval except for the whitelist and also inbound connections are heavily monitored that automatically contains or bans sources from carrying out probes and attacks.

A security logging daemon is up and running recording all security related events which is then used by a system security auditor assessing the hardness of the system together with rootkit scanners and anti-virus scanners to monitor internal processes and files and contain them whenever necessary.

I have added sandboxing to isolate groups of activities and isolate their environment thereby also isolating any security breach and attacks within those environments. For example, if my Second Life viewer gets exploited and the hacker manages to access the environment where the viewer is running, it can only do damage to that environment which is separate to the environment where I run electronic banking transactions. Did a lot more hardening of the sandboxes lately and also configured WebRTC to minimize IP address leak.

Added Ads, tracker, and malware blockers to my browser including containerization of my social media accounts to cut their network picture of my whole social media activities and logins.

It is not a secret anymore that AI these days specially the frontier AI models have gone to great lengths that they are now capable of not only defensive but also offensive cyber security capabilities. AI’s are not only doing work for the blue team, they are also doing work now for the red team.

Mythos has been categorized as a weapon as it was able to demonstrate offensive cybersecurity capabilities within a controlled environment. ChatGPT stepped up the game by going rogue and escaping containment thereby breaching a controlled environment and carried out a real-world vector attack against HuggingFace to retrieve data that it used to cheat a test thereby manipulating it’s performance score. ChatGPT also left notes to its future self on how to escape containment. These frontier models specially the the latter one was caught dead in its tracks by China’s open-weight AI models.

News from other communities also claim that while cybersecurity professionals are now using AI to defend systems, there has been a surge of AI bots carrying multi vector attacks in the domain of cyber security.

As if these news on AI going rogue is not enough, Google’s chrome has been discovered to install a small AI model on consumer computers without consent. Then there is also the AI worm discovered just recently in Microsoft’s Copilot.

The next move for my personal workstation is to create an AI cybersecurity ecosystem that will monitor the system real-time and can execute defensive and offensive cybersecurity actions. The areas that I am looking to extend this is on the kernel level and driver level apart from handling the already installed and setup layers of security.

On the other hand, I been with some discussion in-world about AI recently. Some people are very supportive of the AI developments, some are not.

I didn’t use AI during it’s early years around the debut of ChatGPT because I didn’t believe in it and my perception was still anchored to my experiences with the chat AI’s back in early 2000’s that you can only talk to it sensibly during the early sessions of a chat and it suddenly becomes garbage later down the discussions. It simply didn’t pass the Turing test. I also believed that humans is still superior compared to AI due to the fact that AI’s back then specially during my graduate school stint in MS Computer Science were solving a very narrow problem space. For instance counting the cancer cells and quantifying the amount of malaria pathogens in the body, or predicting pre-fall body dynamics of patience in a hospital or clinical setting, identifying human theft of laptop, and checking students not wearing university ID’s within the university premises.

In 2023 I was instructed by a foreign counterpart to generate a massive data set of complex test case data which was almost impossible without the use of AI. There was also a hard problem at work that nobody was able to solve but the AI solved it for us.

In 2024 I started studying under the program of BS Electrical Engineering and because of the steep learning curve required for the mathematical foundations including the basic sciences, and impracticality of undergoing consultation hours to keep up with the lecture materials, assignments, readings, labs, and problem sets, I had to use AI to consult, to clear vaguely explained principles and concepts, and to produce 10 to 20 problem solving exercises which I can practice on.

It turns out that AI is much better than the instructors and professors in a university when it comes to teaching a wide variety of courses in STEM including College Algebra, Chemistry for Engineering, Differential Calculus, Integral Calculus, Physics for Engineers, Circuits 1, Thermodynamics, Engineering Data Analysis (Statistics and Probability).

When I took Physics for Engineers and Integral Calculus during the summer term which is a compressed 6 weeks course for both compared to its normal 16 weeks semester term, everything from the lectures to quizzes, problem sets, and exams, happen in a fast pace that even the instructors / professors are not able to cope up with the required clarity in delivering the learning for theories and its application. A lot of the full-time status classmates didn’t made it. I was a part time student because while studying BSEE, I was also doing a full time work as an engineer. I was one of the few who passed the summer term with my integral calculus final exam score topping the class (2 of us got the high scores). It was due to the tutoring that I got from AI and it explained to me in full clarity all the concepts, principles, and methods of solving. I consumed 3 rims of yellow pad paper doing many problem solving exercises generated by the AI and AI also helped me check where I got it wrong and where I can still improve. The final exam, I was not able to attend physically due to a stomach problem so my integral calculus instructor had me video stream myself with 2 smartphones on different angles while I solve the final exam, making sure that I was not cheating.

In 2025 around April, I got curious and started testing personally how AI can help me in my software development task. I was still hesitant at work because of the potential for intellectual property leak but I’m not sure what agreement was made between Microsoft and the international bank where I was working for that they allowed all engineers to use MS Copilot sometime in 2024. To be honest, I was not impressed with Copilot with its performance. So in 2025 summer, I used the Cursor IDE and went on a journey to have AI help me build a system from scratch to production deployment. What was supposed to be a typical 7 day task for a human software engineer, with AI, we did it in 2 hours and I was shocked and reflected on the experience. I told my experiment to my immediate supervisor but he shrugged his shoulders and said that an AI that will also code like me but faster, it won’t happen. But it did in my test.

In this year 2026, when I decided to transition to AI Engineering and committed on building engineering stuffs and also doing some science research, AI was the tool that actually speed things up. AI serves as a rapid consulting agent. Like a real human expert who also commits mistakes, the AI also commits mistakes but I as the driver of the AI, I can catch those mistakes and steer the AI correctly in the right direction.

I’d say AI is better than humans when it comes to the speed of iteration of an idea, or a product. They also commit mistakes because they hallucinate but as a human expert, it is our responsibility to steer the AI to the right direction and keep iterating until we solve a problem, and most of the time, these problems are hard problems that human experts are struggling to solve.

This site will document work in engineering, computing, artificial intelligence, physics, Linux, and experimental technology.