The Real Talk:
- Why AI assistants that live in your text messages are fun to test and risky to trust with real work
- The study that found roughly 30% of public AI skills had security vulnerabilities, and why installing one is like letting an attacker in through the front door
- How to stop renting your business processes from big SaaS platforms and start owning the ones that make you different
A Closer Look:
Free Range Agents Eric Brown, CISSP and Nick Mellem have been testing Instinct, a text-based AI assistant that can order lunch for the office, brief you on traffic and meetings over your morning coffee, and store your passwords. That last part is where things get interesting. Eric’s advice is simple: be cautious if you are inclined to give it your credit card. Try experimenting with a virtual card that has a spend limit, and keep anything sensitive out.
Alex Bratton ᯅ, CEO and Chief Geek at AIWhy.io and Lextech, leans a little more conservative and calls this category “free range agents.” They’re running somewhere, doing something, with zero visibility into what happens behind the scenes. Great for experimenting. Not ready for real work.
Don’t Run Strangers’ Skills Skill repositories are the new app stores, and they come with the same problems. Alex points to security researchers who reviewed thousands of skills and plugins across the top repositories and found about 30% had vulnerabilities, with some containing outright malicious code. His rule: use external skills for inspiration, never run them. Ten minutes later you have your own version that you know is safe. A malicious skill doesn’t have to break through any walls. It gets access to everything your AI platform can touch. We couldn’t agree more!
Lease the Commodity, Own the Capability This is the core idea in Alex’s new book, Capability Equity. For years, companies bent their business around big SaaS platforms and filled the gaps with spreadsheets. Now the people who know a process best can turn it into reusable AI skills: plain-language instructions plus a thin slice of code, so 1 + 1 always equals 2. Use a skill in chat a few times. Doing it 50 times a day? Build a micro app. Eventually you may find nobody logs into that expensive back-end system anymore. Alex’s warning to SaaS vendors is blunt: if your only value is being the interface to someone’s data, you won’t be around in two years.
Pick up a copy of Alex’s book here: https://a.co/d/06G2jwhI
Apple’s Quiet Privacy Play Alex argues Apple has correctly treated the LLM itself as a commodity, focusing instead on smaller models that run on the device and a verifiable private cloud for bigger tasks. The consumer company has become a privacy benchmark enterprises should be watching, especially compared to wearables and assistants from companies whose business model is your data. The gap: Siri still doesn’t build memory from your conversations.
Bottom Line: AI assistants are getting more capable and more seamless, and that’s exactly why security teams should be paying attention. Test them, because you can’t secure what you don’t understand. But know where your data goes, build your own skills instead of running someone else’s, and own the processes that make your business unique. As Eric put it after aborting a gusty landing near Albuquerque, sometimes you just have to make the right call, in aviation, in cybersecurity, and in adopting new AI tech.
Tune into the full episode to hear about Alex’s Darth Maul grocery run, how he went from idea to published book in two months using his own AI skills, his decade of plain-burger reviews, and Eric’s take on a takeoff from Nepal’s Lukla Airport.
🔗 Ep 99 – Own the Process: AI Skills, Free Range Agents, and Owning Your IP
Listen wherever you get your podcasts – Subscribe to our YouTube channel to stay up to date on breaking cybersecurity news.
Learn more at www.itauditlabs.com
