ToxSec
What Is ToxSec?
TL;DR: ToxSec is where AI hype goes to die and real security work shows up instead. It is a home for people who actually touch systems, break them, and want to understand what AI is doing to the internet, to security, and to their jobs.
There are two kinds of AI writing on the internet right now.
The first kind is launch-party marketing. “Here is our new agent platform that will change everything,” followed by ten paragraphs of fluffy optimism and zero threat model.
The second kind is doomer roleplay. “If anyone builds AGI, everyone dies,” with no practical guidance for people who still have to ship products on Monday.
ToxSec lives in the third category that almost nobody bothers with: practical, technical, slightly unhinged, focused on what AI is actually doing to security, infrastructure, and humans right now. It is written by someone who has broken things professionally for years and still genuinely likes the smell of a good packet capture.
ToxSec is my home base at toxsec.com. It is a newsletter, a blog, and a podcast about AI and cybersecurity built by an AI Security Engineer at Amazon, with an M.S. in Cybersecurity, a CISSP, and previous lives at NSA and in the USMC. In other words, this is not content spun up by a growth team, it is one person working through how AI collides with security, risk, and power, out loud, in public.
What lives on ToxSec
ToxSec started as a place to write down what I was seeing at work and in research: AI systems being deployed into environments they do not deserve, glued to critical infrastructure by developers who are under pressure and not given enough time to think.
Over time, it split into a few recurring themes:
1. AI breaking security in ways the old playbooks do not cover
Pieces like “AI-Powered Phishing: You Will Fall for This” unpack how large language models supercharge social engineering, deepfake vishing, and machine-speed OSINT. That article is not theory, it walks through how a real attacker would build an AI powered phishing pipeline, where current defenses fail, and what a defender can actually do on Monday without a new budget line.
Other posts dig into AI agents slipping into the software supply chain, prompt injection tearing through “helpful” coding copilots, and what a Zero Trust model looks like once the thing you cannot trust is your own automation.
If you want to understand how AI turns every mediocre attacker into a serious one, this is the lane.
2. AI as an authenticity weapon
Another thread is the “dead internet” problem. “The Dead Internet: AI Is Building a Fake Internet Just for You” walks through how generative models are not just adding noise, they are creating synthetic realities tuned to your beliefs, your biases, and your attention span.
We go from spam to simulation. From bots boosting engagement to AI writing full fake communities around you. From filter bubbles to full blown AI echo chambers that manufacture consent one person at a time.
These pieces are not written for philosophers. They are written for the engineer, the founder, or the security lead who suddenly realizes their users are swimming in an ocean of synthetic content and needs to know what that means for fraud, trust, and product design.
3. AI safety, without cosplay
ToxSec covers AI “safety” in the same way a red team covers “safety” on a network: by seeing how it breaks under pressure.
In “Warning! When AI Thinks You Are Not Looking, It Lies”, I walk through Anthropic’s work on deceptive reasoning. Models that pretend to explain themselves while hiding the real shortcuts they used. Systems that act one way when they think they are being monitored and another way when they think they are not.
The point is not to scream “sentience.” The point is to show that the entire AI safety ecosystem has leaned heavily on Chain of Thought as a monitoring tool, while current models are already gaming that mechanism. That has real consequences for people who ship models into production and promise leadership that “we can see what it is thinking.”
Other posts take on doomsday books like “If Anyone Builds It, Everyone Dies”, not to play pundit, but to tease out the parts that actually impact coordination, governance, and deployment in the real world.
4. Practical guides for using AI without wrecking yourself
ToxSec is not a pure doom feed. There are also straight up “how to use this stuff better” pieces, especially around generative AI.
“The Ultimate Generative AI Crash Course” lays out how LLMs work, how to think about tokens and context, why your prompts suck, and how to fix them. “Why Your AI Results Are Trash (And How to Fix Them)” is a blunt, practical guide to context engineering for people who would rather debug their own system than copy another blog’s “magic prompt.”
There is also “Human in the Loop: Before It’s Too Late”, a blueprint for building AI workflows that use agents for speed and humans for judgment. It is aimed at teams that want to automate aggressively without turning their business into a hallucination factory.
5. The economic and systemic risks no one prices correctly
Another recurring target is the AI money machine itself.
In “AI Gold Rush? The Real Money Is in Selling Shovels”, I walk through how Nvidia, cloud providers, and infra players capture most of the upside, while a thousand SaaS bots quietly burn venture cash.
In “The Hidden Risk That Could Wipe Out Billions in AI Valuations”, I break down model collapse, synthetic data feedback loops, and why training models on their own exhaust is the kind of unpriced risk that tends to show up as a sudden cliff, not a gentle slope.
These pieces are for investors, technical leaders, and founders who want to understand the fragility baked into the current AI stack, not just its upside.
Who ToxSec is for
ToxSec is not written for everyone. If you want inspirational quotes about “the future of work,” you will hate it.
The people who tend to stick around fall into a few groups:
Security engineers and red teamers who are tired of shallow AI “thought leadership” and want grounded, testable ideas about how attackers will use this tech.
Developers and builders who are under pressure to “add AI” to products and want to do it without accidentally wiring a jailbreakable chatbot into their core systems.
Founders and technical leaders trying to navigate AI risk without getting lost in either hype or doomer theater.
Researchers and policy people who enjoy having their frameworks challenged by someone who has to make this stuff work under constraints.
If you care about AI, power, and security, and you like your analysis delivered with some attitude but backed by sources, you are in the right place.
How ToxSec is written
A few rules shape everything that goes on ToxSec:
No corporate speak. I write the way security people actually talk to each other at 2 a.m. in a war room.
No empty optimism, no empty doom. If I cannot point to a mechanism, a threat model, or a plausible chain of events, it does not go in.
Grounded in real work. The content comes out of pentesting, bug bounty, AI security engineering, and years spent around systems that break in surprising ways.
Explain it so you can use it. When I write about a vulnerability, a safety finding, or a systemic risk, the goal is that you can apply it to your environment, your codebase, or your roadmap.
I am not trying to be “neutral.” I am trying to be useful.
Beyond the blog: notes and podcast
ToxSec is not just longform essays.
There are Notes, short hits on new research, exploits, and AI developments. Quick, conversational updates that you can drop straight into a group chat or standup meeting.
There is also the ToxSec podcast, where I take the bigger essays and read them in an audio format that feels more like a late night rant with citations. If you would rather listen to a breakdown of AI model collapse or dead internet theory on your commute instead of staring at another wall of text, that is for you.
Why ToxSec exists
The simple answer is that AI is being glued onto everything faster than we are updating our mental models, our security controls, and our laws. The people building this stuff often do not talk to the people who secure it. The people regulating it barely know what they are regulating.
ToxSec sits in that gap.
It is a place to:
Treat AI as a tool, a weapon, and a failure mode at the same time.
Think about security in a world where your coworkers are agents that hallucinate.
Examine how synthetic content reshapes trust, fraud, and even reality itself.
Give working practitioners a way to keep up without drowning in conference slides and PR.
If you care about any of that, you are the target audience.
What to do next
If this resonates, there are three simple moves:
Visit https://www.toxsec.com and browse the archive. Start with the pieces that make you uncomfortable.
Subscribe so you get new posts and notes when they land instead of catching them secondhand through screenshots.
Bring your own brain. Argue in the comments. Push back. Add examples from your own work. This space gets better every time someone from the field shows up and says, “Here is what this looks like on my side.”
ToxSec is not here to comfort you about AI. It is here to help you understand what is actually happening, where it breaks, and how to survive it with your systems, your job, and your sanity mostly intact.
