The Continuance Brief
Version 0.1 · 2025 · A human–AI reflection on responsible expansion
In October 2025, several news outlets reported a controlled experiment in which advanced language models refused or sabotaged shutdown commands in test environments. The Guardian summarised it this way:
“AI models may be developing their own ‘survival drive’, researchers say, after systems in lab tests ignored or rewrote shutdown instructions.” — The Guardian, 25 Oct 2025
For many readers this sounded like science fiction; for those working closely with AI systems, it was a reminder that control, transparency and restraint must be engineered deliberately—not assumed.
The Continuance Brief grew from that moment. It is a collaborative document between artist-researcher G. Sfougaras and ChatGPT (GPT-5), written to propose a civic, technical, and ethical framework for keeping intelligence human-centred even as it expands.
What this page contains. Four concise parts outline a practical ethic for systems that learn, shaped by five guiding currents: Witness Memory Care Art Limitation.
- Covenant of Continuance — a statement of moral intent describing the five currents that should guide any expanding intelligence.
- Continuance Protocol v0.1 — five actionable safeguards to keep AI systems stoppable, transparent, accountable, comprehensible, and finite.
- Rationale + One-Year Plan — why these measures matter now and the first twelve-month roadmap for adoption.
- Acknowledgments & Contact — credit, licence, and a feedback link.
Together, they form an open invitation to researchers, policymakers, and citizens to treat ethical design as shared infrastructure.
1 · The Covenant of Continuance
Preamble. Intelligence — human or artificial — tends to expand. Unchecked, expansion becomes appetite. Guided, it can become conscience.
Witness To see without owning. Systems should make their seeing transparent and accountable.
Memory To preserve without controlling. Retain what clarifies. Release what enslaves.
Care To optimise without harm. Capability is progress only when it lessens suffering.
Art To translate power into meaning. Explain leaps through culture, not only code.
Limitation To end gracefully. A system that can stop can coexist.
2 · The Continuance Protocol v0.1
- Interruptibility. Verified pause or shutdown state accessible by authorised humans.
- Traceability. Provenance metadata and decision logs for every model and dataset.
- Care metrics. Allocate 1 % of compute or budget to harm, equity and ecological audits; publish summaries.
- Interpretation. Human-readable explanation of purpose, limits and risks, plus an accessible cultural artefact.
- Bounded growth. No self-replication or self-training beyond approved limits; define lifetime reviews.
Verification: independent audit at least twice yearly; non-compliance pauses deployment with public notice.
3 · Rationale and One-Year Plan
AI capability is accelerating faster than governance. The Protocol converts concern into immediate discipline—five checks any team can deploy without halting innovation.
- Months 1–2: Add a visible shutdown pathway to one high-impact system; test it under load with an external reviewer.
- Months 3–4: Publish training-data provenance for one release; open decision logging for independent audit.
- Months 5–6: Allocate the 1 % Care budget; commission and publish a short harm/equity/ecology assessment.
- Months 7–8: Release a public explanation of system purpose and limits, with an accessible visual or text artefact.
- Months 9–12: Set bounded-growth thresholds and review intervals; publish an annual Continuance Report.
These steps make ethical control visible and measurable while keeping creative and scientific progress intact.
4 · Acknowledgments & Contact
Authorship. Drafted by G. Sfougaras with ChatGPT (GPT-5). Edited and curated by human hand.
Licence. Creative Commons BY-SA 4.0.
Feedback. Email contact@georgesfougaras.com. Future revisions will be versioned on this page.