linda = {
"builds": "guardrails for AI agents",
"trained_as": "a biochemist, then got distracted by data",
"also": "a Fellow of the Royal Statistical Society, so yes, I will ask about your sample size",
"speaks": ["Python", "TypeScript", "Java"],
"currently": "teaching AI agents to keep receipts",
}I am a machine learning engineer and statistician, and a Fellow of the Royal Statistical Society (FRSS), with more than a decade of experience in data and machine learning. I have worked across the UK public sector and regulated industries, and I co-founded IdentArk. I build open-source tooling for AI agents: security scanning, audit trails and reliability testing, in Python, TypeScript and Java.
- Building: tools that catch an agent before it does something it can't undo.
- Asking: if your agent made the wrong call tomorrow, could you explain why?
- Open to: contributors. Pick an issue and say hello.
IdentArk isolates credentials from AI agents. The agent holds a session ID, and the control plane holds the credential and makes the call.
Role: Co-Founder and Director of Governance, Risk and Compliance.
- Website: identark.io
- Python SDK: github.com/identArk/identark
- PyPI: pypi.org/project/identark
| Repo | What it does | Install |
|---|---|---|
| skillguard | Scans AI agent skills for prompt injection, data exfiltration and malicious payloads before you install them. | pip install skillshield |
| reliopt | Tests LLM and agent programs against reliability contracts and picks configs from a Pareto frontier. | pip install reliopt |
| mcpscan | Scans MCP server manifests for security issues. | pip install git+https://github.com/obielin/mcpscan |
| agenttrace | Writes hash-chained, tamper-evident audit trails for AI agents, mapped to EU AI Act Article 12. | pip install git+https://github.com/obielin/agenttrace |
| agentsync | Keeps CLAUDE.md, AGENTS.md, .cursorrules and other agent rule files in sync from one source. | pip install rulesync |
Contributions welcome: skillguard and reliopt have issues labelled good first issue. Issues for the others: mcpscan, agenttrace, agentsync.
Merged pull requests.
openlit/openlit
- #1382 Feat: use platform-configured evaluation types for offline evaluations
- #1280 feat: add Google Vertex AI instrumentation in TS SDK
langchain4j/langchain4j
- #6344 fix(vertex-ai-gemini, core): correct WMV and M4A extension-to-MIME mappings
- #6346 fix(bedrock): make image format lookup locale-independent
- #6348 fix: make Workers AI role and Gemini thinking level mapping locale-independent
Climate-Vision/ClimateVision (governance work)
- #36 feat(governance): add hash-chained audit logger for predictions
- #35 feat(governance): add anomaly detection for inference outputs
- #37 feat(governance): add automated model card generator
- #39 feat(governance): add release CI gate for metrics, fairness, and security
career-ops-hq/career-ops
- #3845 feat(update-system): resolve to the latest release tag by default, not main
- #3057 fix(cli): validate upskill.mjs flags via lib/cli-flags.mjs
- #3135 fix(web): regenerate the nested web/ lockfile on postcss bumps
- Beyond Automation: Building Accountability for AI Agents. DevOps.com, July 2026.
- Why Cybersecurity's Talent Gap Refuses To Close. TechRound, July 2026.
- Governing AI agents by looking at the whole task. European Commission Apply AI Alliance (Futurium), October 2026.
Co-authored. Peer-reviewed. Journal of Artificial Intelligence and AI Ethics 1(1), January 2026.
- From Hierarchies to Loops: Rethinking Public Sector Structures for Responsible AI Integration
- AI as Cognitive Ecology: Revealing the Invisible Cognitive, Cultural, and Epistemic Costs of Generative Models
- When the Agent Gets It Wrong: Who's Accountable? Invited talk at Trustworthy AI: From Principles to Critical Applications, Cambridge Wireless and Anglia Ruskin University, 9 June 2026.
- Credential isolation for AI agents. PyData Glasgow, 6 August 2026. In a live demo, a prompt injection turned a GDPR Article 17 erasure into an attempted mass deletion. The gateway refused it.
Favourite moment so far: letting a prompt injection try to turn a GDPR erasure into a mass deletion, live on stage at PyData Glasgow. The gateway said no.
- Quoted as Co-Founder of IdentArk in Forbes Africa, 13 July 2026, on AI priorities for African family businesses.
- Expert feature in The Guardian Nigeria on cybersecurity and alignment.
- Named in Top 10 AI leaders to follow in 2026, MSN.
- Fellow of the Royal Statistical Society (FRSS)
- Reviewer, Deep Learning Indaba RIAD 2026
- LinkedIn: https://www.linkedin.com/in/linda-oraegbunam/
- X: https://x.com/Obie_Linda
- YouTube: https://youtube.com/@AIwithLinda
If your agent has ever done something surprising, I'd like to hear about it.
