What I'm doing
AI Evaluation & Safety contract work. Ongoing red-teaming and adversarial-testing engagements against frontier LLMs on independent platforms. Most of it is under NDA, but the methodology lives in my AI Red Teaming Frameworks case study: jailbreak taxonomy, prompt-injection testing framework, automated adversarial suite, and multimodal attack-surface analysis.
In parallel, I'm actively job-searching for full-time roles where the AI-security and traditional-SOC sides of my work both get used. The thesis I keep coming back to: I bring enterprise security rigor to AI safety, and AI-native thinking to security operations. Most postings ask for one side or the other; the strongest fit is the role that wants both.
What I'm building
Three things are getting most of my evening / weekend cycles right now:
- Adverse Insight, a 3-agent contract risk analyzer (Streamlit + OpenAI) that runs Extractor → Adversarial Scorer → Negotiator passes over a contract draft. Live demo. Currently iterating on the scorer's calibration and adding edge-case test contracts.
- AI Red Teaming Frameworks, the methodology repo that backs my red-team contract work. Keeping the jailbreak taxonomy current as new attack patterns appear (agent-abuse vectors and tool-use exploitation are the active frontier right now). The Pattern Detector and the Red Team Atlas both run on it directly.
- The Lab: chimaukachukwu.com/lab, four interactive AI security exhibits that run entirely in the browser. A prompt injection sandbox where you attack an agent and then defend it, a six-level CTF, a live jailbreak classifier, and the taxonomy as browsable reference. All deterministic. No model call, no backend, no API key. The constraint turned out to make them better: reproducible results mean you can toggle one control and see exactly what it changed.
- This portfolio, treated as a real product rather than a page. Zero runtime dependencies, two author-time generators, CI that fails on stale generated markup or a broken link, and WCAG AA verified with a compositing-aware contrast sweep rather than by eye. The reasoning is written up in the colophon, including where it is still weak.
What I'm learning
Two threads, deliberately kept in parallel because I think they sharpen each other:
- Frontier AI red team. Newer attack surfaces: agent-abuse and tool-use exploitation, multimodal prompt injection (image + text composite payloads), persistent-memory attacks on stateful agents. The taxonomy I maintain has had to grow to accommodate these; the regex-pattern library lags the structural categories on purpose.
- SOC and detection depth. Splunk SPL practice, detection-engineering thinking (writing the kind of rules my Hobby Lobby MISP work would have needed), AWS and Azure security configurations, and getting fluent in tools I touched at the corporate level but didn't own. The CCEP and CTIGA pathways from Red Team Leaders have been useful pressure here.
The bet is that fluency in both sides, plus the ability to translate between them, is a more durable skill than depth in either one alone. Every model upgrade obsoletes specific exploits; structural thinking about attack-and-defense doesn't.
What I'm open to
I'm actively interviewing for the following kinds of roles, weighted roughly evenly:
- AI Security Analyst / AI Red Teamer (offensive or evaluation-focused)
- SOC Analyst (Tier 1 / 2) at organizations that take AI threats seriously
- GRC / AI Governance roles (NIST AI RMF, ISO 42001, OWASP LLM Top 10 territory)
- Hybrid roles that explicitly span both sides, the "bridge" hire
U.S.-based remote, hybrid (Oklahoma City metro), or on-site for the right team. Green card holder, eligible for Secret-level clearance (don't currently hold an active one).
If anything here sounds like a fit for your team, or you'd just like to chat about AI red teaming, the AI-security/SOC bridge, or any of the projects above, reach out. The fastest channel is email: chima.ukachukwu.sec@gmail.com.
About this page
This is a /now page, a convention started by Derek Sivers: a single page that summarizes what you're currently focused on, in lieu of having to explain it in every reply to "how have you been?" If you want a sense of where I'd be if we ran into each other this month, this page is the answer.