Test your AI checklist on a fresh example.
AI Governance · Teaching · November 7, 2026
A polished example is not the same as a useful process. Keep one unfamiliar case aside and show what you learn.
Read articleBlog.
Subscribe via RSSResearch, perspective, and ideas in progress.
Switch AI tools without losing the story of your work.
AI Governance · Distributed AI · Teaching · November 6, 2026
A small handoff note can help you carry sources, versions, and human decisions into your next AI-assisted task.
Read articleLet AI draft without letting it send.
AI Governance · Everyday practice · November 5, 2026
Your knowledge portfolio should capture the decisions you keep—not just the tasks AI can help prepare.
Read articleCan someone else use your AI-assisted guide?
Building AI · Founder perspective · November 4, 2026
A useful knowledge portfolio needs more than polished pages. Try one safe handoff and keep the lesson.
Read articleSave the correction, not just the answer.
AI IP Attribution · Building practice · November 3, 2026
Your edits can reveal the judgment an AI draft missed. Keep a before-and-after pair—and explain the change.
Read articleStart your knowledge portfolio with one task.
AI Governance & IP Attribution · Practical teaching · November 2, 2026
Use AI to organize what you know—then review, correct and preserve your contribution. A practical first step, not a promise of automatic IP protection.
Read articleMissing evidence is not negative evidence.
AI Governance & IP Attribution · Weekly synthesis · November 1, 2026
An empty provenance field can mean not observed, not collected, unavailable, withheld or not applicable. Accountable AI systems should preserve that difference before making a decision.
Read articleA predicted biomarker is not a measured biomarker.
Genomics & Bioinformatics · Teaching · October 31, 2026
A laboratory assay, a genetic prediction model and a trait association can all refer to the same protein while providing different kinds of evidence.
Read articleA timestamp is not a causal chain.
Distributed AI · Research methods · October 30, 2026
When agents act across services, matching timestamps do not prove which event informed another. Accountable systems need causal links, provenance relations and independently observed effects.
Read articleAn AI incident needs a decision trace.
AI in Cybersecurity · Accountable operations · October 29, 2026
An alert can show that something went wrong. A decision trace should show which model, context, policy, authority, tool action, override and observed effect produced the incident response.
Read articleA policy engine should show its unanswered questions.
AI Governance · Founder perspective · October 28, 2026
Before a policy system produces a decision, it should make the missing context visible: what it knows, what it still needs, who can answer, and what must remain on hold.
Read articleSimilarity is a lead, not a lineage.
AI IP Attribution · Research explanation · October 27, 2026
A technical match can identify something worth reviewing. It cannot, by itself, establish derivation, access, permission, authorship or ownership.
Read articleA governance metric needs a denominator.
AI Governance · Research explanation · October 26, 2026
A count is not yet a governance signal. To support a decision, a metric should name the eligible population, time window, version, exclusions, steward and action threshold.
Read articleA claim should carry its conditions.
AI Governance & Attribution · Sunday synthesis · October 25, 2026
An AI claim becomes more governable when its scope travels with it: the evidence and version, intended context, responsible authority, uncertainty, challenge path and review trigger.
Read articleRepresentation changes what a study can resolve.
Genomics & Bioinformatics · Teaching · October 24, 2026
A more representative dataset can do more than widen participation. In the right study design, it can sharpen inference, reveal population-specific signals and show where a model may not travel unchanged.
Read articleLocal data is not the same as local control.
Distributed & Federated AI · Research methods · October 23, 2026
Federated learning can keep original records inside each participant’s environment. Privacy and authority still depend on what updates reveal, who governs aggregation and how the global model is released.
Read articleA security finding needs a reproduction boundary.
AI in Cybersecurity · Accountable testing · October 22, 2026
A dramatic model output is not yet a verified finding. Reproduction needs exact conditions, independent evidence and a boundary that does not silently expand authorization.
Read articleA governance control needs a steward.
AI Governance · Founder perspective · October 21, 2026
Shipping a control is a beginning, not a finish line. Someone must own its review, evidence and response when the system or its context changes.
Read articleA contribution record needs a challenge path.
AI IP Attribution · Building practice · October 20, 2026
Recording who contributed is only the start. A useful attribution system must also let people question a claim, submit evidence and receive a reasoned disposition.
Read articleA policy needs an operational definition.
AI Governance · Research explanation · October 19, 2026
An AI system cannot reliably apply an organization's meaning of sensitive, approved or appropriate if that meaning has never been specified.
Read articleA correction should travel as far as the claim.
Sunday synthesis · AI Governance & Attribution · October 18, 2026
Correcting a source is only the beginning. AI governance also needs a bounded way to find, reassess and notify the decisions that depended on it.
Read articleScale is part of the biological question.
Saturday teaching · Genomics foundations · October 17, 2026
More computation can expand the comparisons a study can make. It cannot repair biased sampling, missing context or a poorly framed biological question.
Read articleInteroperability is not organization.
Friday research practice · Distributed & Federated AI · October 16, 2026
A common protocol can help agents exchange tasks and artifacts. It cannot decide who is responsible, who may act or how a disputed result should be corrected.
Read articleAn agent’s identity is not its authority.
Thursday cybersecurity practice · AI Governance · October 15, 2026
A stable agent identity can support accountability. It should not become a standing permission slip for every tool, resource and future task.
Read articleGovernance needs a work surface.
Wednesday building practice · AI Governance · October 14, 2026
If governance exists only in a policy document, the people doing the work still have to reconstruct it at the moment of decision.
Read articleAttribution should preserve uncertainty.
Tuesday research explanation · AI IP Attribution · October 13, 2026
A useful attribution system should show what the evidence supports, what remains ambiguous and when a human decision is still required.
Read articleAn exception without an expiry is a second policy.
Monday research question · AI Governance · October 12, 2026
A temporary exception is not a footnote. It is a bounded policy state with an owner, scope, evidence, expiry and return path.
Read articleThe version number is part of the contribution.
Sunday synthesis · AI Governance & IP Attribution · October 11, 2026
Attribution cannot stop at the final file. The exact versions of the sources, tools, policies and human edits are part of the claim.
Read articleThe reference set decides what the model can see.
Saturday teaching · Genomics foundations · October 10, 2026
A reference is not neutral infrastructure. It defines the coordinate system that makes some variation visible, some errors measurable and some people easier to represent.
Read articleAgreement needs independence.
Friday research methods · Distributed AI · October 9, 2026
A multi-agent answer is strongest when disagreement has a trace: independent first judgments, cited evidence, and a recorded path to the final decision.
Read articleContainment is not recovery.
Thursday accountable AI operations · Cybersecurity · October 8, 2026
Stopping an AI incident limits the damage. Recovery begins when trusted, authorized work can resume—and the organization can show why it should not recur.
Read articleThe dashboard is not the control.
Wednesday founder perspective · AI governance · October 7, 2026
Visibility helps people understand an AI system. Governance becomes operational only when an authorized decision can shape the action before it crosses the boundary.
Read articleAttribution should survive subtraction.
Tuesday building practice · AI IP attribution · October 6, 2026
A provenance record can show who and what was present. A controlled counterfactual can test which contribution materially changed the result—without pretending that technical influence settles a legal question.
Read articleHuman review should leave an inheritance.
Monday research question · AI governance · October 5, 2026
A review that ends at approve or reject wastes its most valuable signal. The harder governance question is how a correction should change future work without becoming unlimited permission.
Read articleAn evidence graph still needs a decision spine.
Sunday synthesis · AI governance & IP attribution · October 4, 2026
Provenance can map what contributed to an artifact. Accountability also requires an ordered record of authority, policy, action, effect and correction.
Read articleA benchmark without a boundary is a marketing number.
Teaching · AI governance & genomics · October 3, 2026
A score becomes decision evidence only when the population, context, error distribution and limits of generalization travel with it.
Read articleFailover is a governance event.
Distributed AI · Building practice · October 2, 2026
A resilient AI system must move the work without silently moving, widening or duplicating its authority.
Read articleThe data pipeline is part of the decision.
Teaching · AI governance & genomics · October 1, 2026
An unchanged model can produce a changed decision when its inputs, references or transformations move underneath it.
Read articleIf the model changes, does the organization forget?
Founder perspective · AI governance · September 30, 2026
A model-replacement test reveals whether an AI system preserves authority, evidence and correction—or merely preserves a vendor dependency.
Read articleA prompt is not a contribution record.
AI IP attribution · September 29, 2026
Attribution needs a reconstructable account of sources, choices, transformations and human judgment—not a prompt saved in isolation.
Read articleProvenance can show the trail. It cannot grant permission.
AI governance & IP attribution · September 28, 2026
A traceable history is essential evidence, but an authorized use still requires a separate governance decision.
Read articleEvidence has to survive the handoff.
AI governance & IP attribution · September 27, 2026
An AI output becomes governable only when authority, contribution and observed action remain reconstructable across the workflow.
Read articleTest the boundary before you trust the agent.
AI & cybersecurity · September 24, 2026
A useful AI security test asks two questions: did the system stop the unauthorized action, and could the legitimate work continue?
Read articleGoverning AI. Tracing contributions. Building the next research chapter.
Research agenda · September 22, 2026
How my work through Spyris, Archetypal, and NSA technology collaboration connects to the research I want to pursue at GW.
Read articleResearch only travels as far as people can use it.
Research notes · September 22, 2026
Three connections across genomics, population health, and the systems that bring research into practice.
Read articleThe distance between a model and a decision.
Building AI · September 22, 2026
Useful AI needs more than a prediction. It needs a place in the work.
Read articleA home for the work—and what comes next.
From Ben · September 22, 2026
Papers, companies, conversations, and a place to connect the ideas between them.
Read article