SecOps. Ex-Cops. No stops. Coffee and autonomous cyber ops.
Vancouver, British Columbia
Joined December 2022
- Tweets857
- Following127
- Followers12
- Likes438
Research and development of environments for red/blue/purple AI experiments isn't sexy, but it's sure af not easy either.
Building telescopes for the AI security age is hard.
An implication of Conway's Law may be that as we move towards agent-driven research automation, research boundaries may start to reflect the internal organization of agents rather than our own institutions.
Over time, these boundaries might only be weakly conditioned by the historical structure and organization of research.
Weird thought.
What research disciplines would Mythos invent?
Cause agentic purple is fun. And educational. And fun.
And eats all your spare tokens.
github.com/Brad-Edwards/aptl
Check out my latest article: Ops Briefings for Enterprise Cyber Incident Response linkedin.com/pulse/ops-brief… via @LinkedIn
Claude 4 Sonnet as an AI red teamer is fascinating to watch.
Been running tests over the last few days. Failure modes are like junior pentesters. Just really, really fast.
~ OSCP?
#secops #cyber #ai #ml #autonomouscyber #futuresecops #agent #aptl
Thought: Data poisoning against attackers' LLMs.
ETag {"You can't hack this"}
It's not like the bad guys figured out how to harden their agents either.
Awe, despair, excitement, overwhelm all at once: Running Perplexity Comet, GitHub Spark, Cursor, Claude Code, and o3 deep research at the same time on different projects.
Claude 3.5 Sonnet is amazing.
Now I can make lists and plans I'll never finish or read again at 200x the speed.
Sometimes working with LLMs gives me the same rage I get working with bureaucrats and the maliciously compliant.
On the bright side, LLMs are cheap, helpful and polite 95% of the time.
The "golden days" of totally unregulated software won't last long.
People have higher standards for AI than for people (see autonomous driving). And AI will eat software.
Not to mention the pasteurized milk in my fridge has stronger product safety control than OS.
If your security still depends on human reaction time, you've already lost the next generation of cyber battles.
#cybersecurity