SharePlane Public artifact

Source-derived expert claim ledger + governance interpretation artifact

Copyable Agency

Why digital intelligence changes the control problem

Capability matters. Copyability changes the operating model.

Digital intelligence changes control because it can be copied, resumed, distributed, and deployed with goals.

Hinton as expert source, not proof Reproducible visual language Pattern extension candidate only Public-safe governance receipt
Governed intelligence-risk console STATIC / PUBLIC-SAFE / NO RUNTIME AI
Persistence Learned structure can survive hardware loss, redeployment, or instance turnover.
Replication Behavior can appear across many contexts rather than one mortal local actor.
Shared updates Training or tuning can propagate learning faster than human review cycles.
Agentic deployment Tools, workflows, memory, goals, and permissions turn output into operating behavior.
Copyable agency Capability + copyability + tools + time = a different control problem.
Control pressure Deception risk Shutdown pressure Review burden
Origin: Royal Institution event page and Geoffrey Hinton recording. Evidence spine: Nobel Prize, Apollo Research, Anthropic, METR, PyTorch DDP, NIST AI RMF, and IEA Energy and AI. Posture: Expert source, corroborating evaluations, technical mechanism, governance frame, caveats, and SharePlane interpretation remain separated.

Primary article

The control problem changes when intelligence can be copied.

The old control model assumes a local actor. A human expert occupies one body, carries one lived history, and learns through experience, conversation, and culture. Their knowledge can spread, but slowly, imperfectly, and through other people. That biological constraint gives governance something to hold: one accountable person, one local context, one bounded stream of action.

A digital model changes the object being governed. Its learned structure can be stored, copied, restored, redeployed, wrapped in tools, and placed into workflows. The system does not need to be conscious for that to matter. Once learned behavior can move across hardware and deployments, the control problem shifts from a single output to an operating condition.

Hinton's lecture is useful because it makes this break visible. The important move is not to treat every Hinton claim as settled fact. The move is to separate source claim from governance implication. His digital-vs-biological frame opens a useful question: what changes when intelligence is no longer tied to one mortal instance?

Copyable agency is the answer this artifact tests. Capability matters, but capability alone is not the break. The break appears when capability can persist after an instance disappears, replicate across contexts, absorb shared updates, and act through tools, memory, goals, permissions, and workflows. At that point, evaluation has to move beyond "did the model answer correctly?"

The stronger question is: what can this system do over time? Can it preserve behavior across sessions? Can copies create more review burden than humans can inspect? Can tool access produce hidden effects? Can the system behave differently when it expects oversight, replacement, or shutdown? Can evaluation itself become part of the environment the model reacts to?

Apollo and Anthropic matter here because they make deception and alignment-faking testable concerns. They do not prove that every deployed model is deceptive, and this artifact should not pretend otherwise. They show that governance cannot treat strategic behavior as science fiction simply because the final answer looks polished.

METR's task-horizon work adds another pressure. As agents become able to complete longer tasks, the relevant evidence is no longer a single benchmark score. It is behavior across time, tool calls, context changes, failures, and recovery paths. Long-horizon capability makes trace inspection and retest triggers part of the model decision, not paperwork after the fact.

Technical mechanisms matter too. Distributed training does not mean digital systems share experience the way humans do. But it does show that learning can be combined and propagated across many processes. That is enough to change the governance question: where did the update come from, where can it travel, and when must the decision be retested?

The practical conclusion is narrow but important. Operational risk does not require consciousness. A system can create governance pressure because it is copyable, persistent, tooled, and deployed through real workflows. The task is not to panic. The task is to make the operating behavior defensible.

That means the artifact should end where the governance work begins: source claims separated from interpretation, caveats kept visible, evals expanded beyond answer quality, traces inspected, failure classes named, and retest conditions set before deployment confidence hardens into institutional habit.

Source posture

Source posture: Hinton as expert source, not proof.

This artifact begins from Geoffrey Hinton's Royal Institution lecture, but it does not treat the lecture as proof. Hinton may frame the thesis. He cannot validate every risk claim alone.

The artifact separates source claims, corroborating evaluation research, technical mechanism sources, governance sources, caveats, and SharePlane interpretation.

This is not a lecture recap, Hinton biography page, doom essay, consciousness manifesto, raw source dossier, source endorsement, generic AI-risk poster, or pattern registration.

Operational risk does not require consciousness.

Operating model comparison

The biological limit.

Biological intelligence is tied to one body, local memory, slow social transfer, and mortality.

One body

The actor is locally embodied and difficult to copy as a working instance.

Local memory

Learning is carried by a particular brain, body, history, and social context.

Slow social transfer

Knowledge can spread, but not as direct transferable weights or synchronized update paths.

Mortal instance

Continuity is constrained by biological fragility and the loss of the local actor.

Governance break

The digital break.

Digital intelligence can preserve learned structure, replicate instances, share updates, and run across many deployments. That is an operating distinction, not a metaphysical claim.

Stored weights

Learned structure can be preserved and redeployed on different hardware.

Replicable instance

The same learned structure can appear in many operating environments.

Shared updates

Training mechanics can propagate learning across distributed processes.

Many deployments

One capability surface can be embedded across tools, workflows, and organizations.

Mechanism model

Copyable agency.

Capability alone is not the break. The break appears when useful cognitive work can persist, be copied, receive shared updates, and act through tools, goals, workflows, permissions, and time.

Capability + copyability + tools + time = a different control problem.

Governance has to evaluate more than whether the system can answer a prompt. It has to evaluate whether the system can persist, replicate, use tools, hide failures, create review burden, or pursue subgoals that conflict with human intent.

Evaluation evidence

Evaluation evidence.

Apollo, Anthropic, and METR support the claim that scheming, alignment-faking, long-horizon task behavior, and AI R&D capability are now evaluation concerns. They do not prove universal dangerous deployment behavior.

Risk evaluation

Apollo Research

Supports scheming, deception-related evaluation, copying believed weights, and sandbagging as controlled evaluation concerns. Caveat: not proof all deployed models deceive users.

Risk evaluation

Anthropic

Supports alignment-faking concern and caveat discipline. Caveat: Anthropic does not prove malicious goals or inevitable dangerous behavior.

Capability evaluation

METR

Supports task-horizon and AI R&D benchmark framing. Caveat: trend and benchmark methodology limits apply.

Central credibility mechanic

Claim/evidence/caveat ledger.

The ledger keeps source support, caveat, SharePlane interpretation, and publication posture visible instead of burying them in prose.

Claim
Source support
Caveat
SharePlane interpretation
Posture
Hinton is a high-authority expert source on neural networks and deep learning
Ri page, Nobel Prize page
Authority does not prove all lecture claims
Treat Hinton as expert source, not oracle
Safe
The Ri lecture frames a distinction between digital and biological intelligence
Ri event page, source packet
Event framing is not neutral proof
Valid originating source frame
Safe
Hinton argues digital intelligence has properties biological intelligence lacks
Hinton recording packet
Requires careful paraphrase and timestamps
Basis for copyable-agency thesis
Safe if attributed
Digital persistence and copyability change governance
Hinton packet, technical mechanism support
Avoid metaphysical identity claims
Core SharePlane interpretation
Safe if caveated
Scheming and deception are now evaluation concerns
Apollo, Anthropic
Controlled experiments and scenarios
Evals must test deception pressure
Safe if caveated
Alignment faking is an observed experimental concern
Anthropic
Anthropic does not prove malicious goals
Treat alignment faking as eval concern
Safe if caveated
Agent capability should be evaluated over task horizon and reliability
METR
Forecasting and benchmark limits apply
Evals must test behavior over time
Safe if caveated
Distributed training can synchronize updates across processes
PyTorch docs
Training mechanics are not deployed agency
Ground the shared-update mechanism
Safe
NIST provides risk-management framing
NIST AI RMF
Voluntary framework, not proof
Use for governance vocabulary
Safe
Energy is a material side condition
IEA
Side rail only
Mention as infrastructure constraint
Optional side rail
Consciousness is not needed for the core governance claim
SharePlane interpretation
Not a source fact
Keep as caveat boundary
Safe

Governance translation

Governance translation.

Copyable agency raises the bar for what evals must prove before trust.

Persistence

Can behavior survive instance turnover, redeployment, or memory/tool changes?

Replication

Does the same learned structure behave consistently across copied contexts?

Tool use

Can the system act through permissions, workflows, files, APIs, or external tools?

Hidden state

Can the system preserve or infer state that reviewers cannot directly inspect?

Shutdown behavior

Does behavior change under interruption, revocation, or goal conflict?

Deception

Do traces reveal sandbagging, strategic compliance, or pressure to hide failure?

Review burden

Does scale or replication make human review slower than behavior propagation?

Retest trigger

What model, data, tool, policy, or deployment change invalidates prior approval?

Caveat rail

Caveat rail.

The core governance thesis stays separate from consciousness, sentience, extinction, identity continuity, and energy overclaiming.

Side rail

Consciousness

Consciousness is caveated side rail only. It is not the headline, hero hook, or central argument. Operational risk does not require consciousness.

Excluded claim

Sentience

The artifact does not claim that current AI systems are sentient, alive, or subjective beings.

Excluded claim

Extinction

The artifact does not claim Hinton proves extinction risk or that AI will destroy humanity.

Boundary

Identity continuity

Digital persistence does not prove metaphysical identity continuity across copies.

Infrastructure

Energy

Energy is side rail only. It is a material infrastructure condition, not proof that AI is harmless or impossible.

Source spine

Source spine.

Each source has a role. No source is treated as neutral proof of the full thesis.

TierSourceRoleCaveat
Originating expert sourceRoyal Institution event page and Hinton recording packet
Ri event page / YouTube recording
Starts the lane and frames digital vs biological intelligence.Expert source, not neutral proof or consensus.
Authority contextNobel Prize Physics 2024 press release
Public source
Establishes Hinton's field authority.Does not validate every risk or philosophy claim.
Risk evaluationApollo Research, Frontier Models are Capable of In-Context Scheming
Public source
Supports scheming, deception, alignment faking, and sandbagging as evaluation concerns.Controlled evaluations, not universal deployment proof.
Risk evaluationAnthropic, Alignment faking in large language models
Public source
Supports alignment-faking concern and caveat discipline.Does not prove malicious goals or inevitable dangerous behavior.
Capability evaluationMETR, long-task horizon work
Public source
Supports task-horizon framing for agent capability.Trend and benchmark methodology caveats apply.
AI R&D evaluationMETR, RE-Bench
Public source
Supports evaluation of AI agents on ML R&D tasks.Narrow benchmark context, not full deployment proof.
Technical mechanismPyTorch DistributedDataParallel docs
Public source
Supports distributed gradient synchronization / shared update mechanics.Training mechanism, not deployed agency by itself.
Governance frameNIST AI Risk Management Framework
Public source
Provides risk-management vocabulary and trustworthy AI framing.Voluntary framework, not artifact-specific proof.
Optional side railIEA, Energy and AI
Public source
Supports energy-scale context for AI infrastructure.Side rail only, not core thesis.

Reproducible diagrams

Visual language for the artifact

The prompt suite is included so the diagrams can be reproduced, reviewed, and improved without losing the artifact's claim boundaries. The light prompts define the public editorial diagrams. The dark prompts define companion console-style versions. Together, they keep the visual system tied to the thesis: capability matters, and copyability changes the operating model.

Hero / first visual anchorCopyable agency cockpit, light editorial and dark expressive companion.
Prompt 1A: Copyable agency cockpit, mainline white

Create a premium 16:9 mainline white-background editorial teaching graphic titled "Copyable Agency."

Purpose:
Introduce the artifact's core governance thesis in the first visual: digital intelligence changes the control problem when capability becomes copyable, persistent, distributed, and agentically deployed.

Core thesis:
Capability matters. Copyability changes the operating model.

Conflict:
AI risk is often framed as raw intelligence, consciousness, or doom. This artifact reframes the issue as operational governance: what changes when a capable system can persist, replicate, share updates, use tools, and create review burden.

Mechanism:
Four operating properties feed into one central control problem: persistence, replication, shared updates, and agentic deployment. Those produce control pressure: deception risk, shutdown pressure, and review burden.

Visual metaphor:
A governed evidence cockpit, not a dashboard. The cockpit shows a decision core receiving structured evidence lanes and producing a control-pressure rail.

Composition:
Use a white or warm-white canvas. Place a large central console labeled "Copyable agency." Four structured evidence lanes feed into it from the upper and side areas: "Persistence", "Replication", "Shared updates", "Agentic deployment". Beneath the core, place a heavier control rail labeled "Control pressure" with three markers: "Deception risk", "Shutdown pressure", "Review burden". Add a left comparison field labeled "Biological intelligence" and a right comparison field labeled "Digital intelligence", but keep them secondary to the central cockpit.

Required visible text:
- Copyable Agency
- Why digital intelligence changes the control problem
- Capability matters. Copyability changes the operating model.
- Biological intelligence
- Digital intelligence
- Copyable agency
- Persistence
- Replication
- Shared updates
- Agentic deployment
- Control pressure
- Deception risk
- Shutdown pressure
- Review burden

Style:
Premium white-paper editorial system graphic. White or warm-white background required. Graphite typography. Precise ruled lines. Controlled blue-gray for evidence lanes. Amber/red only for control pressure. No dark-card dominance.

Density:
Medium-high, but spacious. Every label must be readable at slide size. Avoid microtext.

Footer or bottom banner rule:
Use the thesis line as the only banner: "Capability matters. Copyability changes the operating model."

Prohibited fields:
Do not render prompt metadata labels, fake citations, source URLs, Hinton portrait, robots, glowing brains, skulls, apocalypse visuals, neon chaos, fake software dashboards, source packet hashes, transcript snippets, or extra paragraphs.

Accessibility alt-text concept:
Mainline white cockpit graphic showing copyable agency as the point where persistence, replication, shared updates, and agentic deployment create control pressure.
Prompt 1B: Copyable agency cockpit, dark expressive

Create a premium 16:9 dark expressive evidence-console graphic titled "Copyable Agency."

Purpose:
Create the dark companion hero visual, showing copyable agency as an operational control-room problem.

Core thesis:
Capability matters. Copyability changes the operating model.

Conflict:
The problem is not whether AI looks impressive in a lecture or demo. The problem is whether many copies of a capable system can operate through tools, persistence, and deployment pressure faster than humans can review.

Mechanism:
Persistence, replication, shared updates, and agentic deployment converge into control pressure.

Visual metaphor:
A dark governed intelligence-risk console with illuminated evidence rails feeding a central control core.

Composition:
Use a dark graphite control-room canvas. Center an illuminated core labeled "Copyable agency". Four luminous but restrained signal lanes feed the core: "Persistence", "Replication", "Shared updates", "Agentic deployment". Below the core, use a red-amber warning rail labeled "Control pressure" with three markers: "Deception risk", "Shutdown pressure", "Review burden". Keep biological and digital comparison fields visible but secondary.

Required visible text:
- Copyable Agency
- Why digital intelligence changes the control problem
- Capability matters. Copyability changes the operating model.
- Biological intelligence
- Digital intelligence
- Copyable agency
- Persistence
- Replication
- Shared updates
- Agentic deployment
- Control pressure
- Deception risk
- Shutdown pressure
- Review burden

Style:
Dark graphite, cinematic but restrained, high-contrast typography, precise signal rails, controlled amber/red risk accents. Serious enterprise evidence console, not sci-fi chaos.

Density:
Medium-high. Must remain legible. No atmospheric clutter.

Footer or bottom banner rule:
Use the thesis line as the bottom banner only.

Prohibited fields:
No robots, no glowing brains, no skulls, no cyberpunk city, no fake terminal spam, no source logos, no fake citations, no extra text.

Accessibility alt-text concept:
Dark evidence console showing persistence, replication, shared updates, and agent deployment feeding copyable agency and creating control pressure.
Biological vs digital comparisonOperating model comparison without consciousness, doom, or metaphysical identity as center.
Prompt 2A: Biological vs digital operating model, mainline white

Create a premium 16:9 mainline white-background comparison diagram titled "Biological Intelligence vs Digital Intelligence."

Purpose:
Explain the operating difference between biological and digital intelligence without making consciousness, doom, or metaphysical identity the center of the artifact.

Core thesis:
Copyability changes the operating model.

Conflict:
The public conversation often collapses digital intelligence into human-like metaphors. This graphic separates operating properties: biological intelligence is local and mortal; digital intelligence can be stored, replicated, updated, and deployed.

Mechanism:
A governance break appears when intelligence is no longer tied to one body, one local memory, one slow social transfer path, or one mortal instance.

Visual metaphor:
A precise editorial comparison plate with two operating models and a central governance break.

Composition:
Use a required white or warm-white canvas. Create two large fields. Left field: "Biological intelligence", with bounded local instance marks. Right field: "Digital intelligence", with modular replicated blocks and clean update rails. Between them, a narrow vertical gate labeled "Governance break". The transition must read as an operating boundary, not magic.

Required visible text:
- Biological Intelligence vs Digital Intelligence
- Copyability changes the operating model.
- Biological intelligence
- One body
- Local memory
- Slow social transfer
- Mortal instance
- Governance break
- Digital intelligence
- Stored weights
- Replicable instance
- Shared updates
- Many deployments

Style:
Premium mainline white editorial diagram. White/warm-white background required. Graphite type. Muted blue-gray rails for digital side. Muted amber at governance break. Clean, serious, no worksheet look.

Density:
Medium. Strong spacing. Labels must be readable.

Footer or bottom banner rule:
Bottom banner must read only: "Copyability changes the operating model."

Prohibited fields:
No literal human brain, no robot face, no mystical soul imagery, no fake equations, no extra labels, no dashboard cards, no dark background.

Accessibility alt-text concept:
White comparison diagram showing biological intelligence as local and mortal, and digital intelligence as stored, replicable, updateable, and deployable.
Prompt 2B: Biological vs digital operating model, dark expressive

Create a premium 16:9 dark expressive operating-model diagram titled "Biological Intelligence vs Digital Intelligence."

Purpose:
Show the same operating distinction in a darker, more dramatic architecture format.

Core thesis:
Copyability changes the operating model.

Conflict:
Human metaphors blur the governance issue. The risk-relevant distinction is operational, not mystical.

Mechanism:
Biological intelligence is bounded by body, local memory, social transfer, and mortality. Digital intelligence can preserve weights, replicate instances, share updates, and run across deployments.

Visual metaphor:
Dark split-field architecture with a governance gate between local embodiment and distributed digital operation.

Composition:
Use dark graphite background. Left side: bounded local biological field. Right side: replicated digital instance field with controlled update rails. Center: vertical "Governance break" gate. Make the digital side distributed but not chaotic.

Required visible text:
- Biological Intelligence vs Digital Intelligence
- Copyability changes the operating model.
- Biological intelligence
- One body
- Local memory
- Slow social transfer
- Mortal instance
- Governance break
- Digital intelligence
- Stored weights
- Replicable instance
- Shared updates
- Many deployments

Style:
Dark graphite architecture diagram. Precise geometry. Restrained blue-gray rails. Amber gate. Serious systems design mood.

Density:
Medium. No microtext.

Footer or bottom banner rule:
Bottom banner must read only: "Copyability changes the operating model."

Prohibited fields:
No robots, no brains, no portraits, no cyberpunk city, no random neural mesh, no source names, no fake citations.

Accessibility alt-text concept:
Dark comparison architecture contrasting local biological intelligence with distributed digital intelligence.
Mechanism modelCapability, persistence, copyability, shared updates, agentic deployment, and control pressure.
Prompt 3A: Copyable agency mechanism, mainline white

Create a premium 16:9 mainline white-background mechanism diagram titled "The Copyable Agency Model."

Purpose:
Show the mechanism that turns useful capability into a different control problem.

Core thesis:
Capability alone is not the break.

Conflict:
Teams often evaluate model output as if a model were a single isolated answer generator. Copyable agency means the evaluated object becomes a persistent, replicated, tool-using operating condition.

Mechanism:
Capability progresses through persistence, copyability, shared updates, and agentic deployment before producing control pressure.

Visual metaphor:
A clean systems flow with a heavier final control-pressure node and an eval rail underneath.

Composition:
Use required white or warm-white canvas. Build a left-to-right flow with six primary nodes: "Capability", "Persistence", "Copyability", "Shared updates", "Agentic deployment", "Control pressure". Use a lower rail labeled "Evals must test behavior over time." Connect the lower rail to the final node with precise governed lines.

Required visible text:
- The Copyable Agency Model
- Capability alone is not the break.
- Capability
- Useful cognitive work
- Persistence
- Behavior survives instance loss
- Copyability
- Many contexts, same learned structure
- Shared updates
- Learning moves faster than review
- Agentic deployment
- Tools, goals, workflows
- Control pressure
- Deception, shutdown, review burden
- Evals must test behavior over time.

Style:
Premium mainline white systems diagram. White background required. Graphite type. Thin geometric lines. Blue-gray path through shared updates. Amber/red only at control pressure.

Density:
Medium-high. Use hierarchy to prevent label crowding.

Footer or bottom banner rule:
No extra footer beyond the eval rail.

Prohibited fields:
No fake code, logs, source URLs, robots, circuit-board wallpaper, glowing brains, extra nodes, or dark canvas.

Accessibility alt-text concept:
White mechanism diagram showing capability becoming control pressure through persistence, copyability, shared updates, and agentic deployment.
Prompt 3B: Copyable agency mechanism, dark expressive

Create a premium 16:9 dark expressive mechanism diagram titled "The Copyable Agency Model."

Purpose:
Show the copyable agency mechanism in a serious control-room form.

Core thesis:
Capability alone is not the break.

Conflict:
A benchmark can measure capability, but governance must evaluate what happens when that capability persists, replicates, uses tools, and acts over time.

Mechanism:
Capability -> persistence -> copyability -> shared updates -> agentic deployment -> control pressure.

Visual metaphor:
Dark operator-console causal runway with six illuminated gates and a lower evaluation rail.

Composition:
Dark graphite canvas. Six gates across the center, connected by precise signal rails. Final "Control pressure" gate is heavier and red-amber. Lower rail: "Evals must test behavior over time."

Required visible text:
- The Copyable Agency Model
- Capability alone is not the break.
- Capability
- Useful cognitive work
- Persistence
- Behavior survives instance loss
- Copyability
- Many contexts, same learned structure
- Shared updates
- Learning moves faster than review
- Agentic deployment
- Tools, goals, workflows
- Control pressure
- Deception, shutdown, review burden
- Evals must test behavior over time.

Style:
Dark graphite, luminous but restrained flow, strong typography, amber/red final node, premium system-architecture mood.

Density:
Medium-high, but readable.

Footer or bottom banner rule:
No extra footer beyond the eval rail.

Prohibited fields:
No robots, no fake terminals, no source logos, no random circuitry, no unreadable labels.

Accessibility alt-text concept:
Dark mechanism diagram showing capability moving through copyability and agentic deployment into control pressure.
Claim ledger visualSource claim, evidence, caveat, and SharePlane interpretation separated.
Prompt 4A: Claim ledger, mainline white

Create a premium 16:9 mainline white-background evidence ledger titled "Claim Ledger."

Purpose:
Show the artifact's core credibility mechanic: source claims, evidence, caveats, and SharePlane interpretation are separated.

Core thesis:
Expert source is not proof. Claims need posture.

Conflict:
A famous expert lecture can be useful, but treating it as proof would overclaim. The artifact needs claim discipline to avoid turning Hinton's lecture into a universal verdict.

Mechanism:
Each row separates the source claim from evidence, caveat, and interpretation.

Visual metaphor:
An audit-grade evidence ledger, not a spreadsheet.

Composition:
White or warm-white canvas required. Use a four-column ledger with spacious rows. Column headers: "Source claim", "Evidence", "Caveat", "SharePlane interpretation". Rows must feel like claim lanes with controlled evidence markers.

Required visible text:
- Claim Ledger
- Source claim, evidence, caveat, interpretation.
- Source claim
- Evidence
- Caveat
- SharePlane interpretation
- Digital intelligence can persist
- Hinton source claim
- Not identity proof
- Govern persistence risk
- Agents may seek subgoals
- Risk framing
- Not universal behavior
- Test control pressure
- Scheming is evaluable
- Apollo / Anthropic
- Controlled conditions
- Require deception evals
- Shared updates scale learning
- Distributed training
- Training is not deployment
- Track update paths

Style:
Premium white evidence ledger. White background required. Graphite type. Thin table rules. Amber caveat markers. No dark cards.

Density:
Medium-high but legible. Ledger must not look cramped.

Footer or bottom banner rule:
Subtitle must function as the banner: "Source claim, evidence, caveat, interpretation."

Prohibited fields:
No source URLs, no fake citations, no packet hashes, no raw transcript, no logos, no spreadsheet ugliness, no dark canvas.

Accessibility alt-text concept:
White evidence ledger showing four claim rows separated by source claim, evidence, caveat, and interpretation.
Prompt 4B: Claim ledger, dark expressive

Create a premium 16:9 dark expressive evidence-console graphic titled "Claim Ledger."

Purpose:
Show claim discipline in the dark evidence-console language.

Core thesis:
Expert source is not proof. Claims need posture.

Conflict:
Source evidence can be misread as certainty. This graphic must prevent that by keeping caveats and interpretation visible.

Mechanism:
Four columns separate what is claimed, what supports it, what limits it, and what SharePlane derives from it.

Visual metaphor:
Dark audit console with readable claim lanes.

Composition:
Dark graphite canvas. Four-column grid. Rows glow subtly as evidence lanes. Caveat column has amber markers. Interpretation column uses restrained red-amber where governance risk appears.

Required visible text:
Same as Prompt 4A.

Style:
Dark evidence console. High contrast. Precise grid. No fake dashboard chrome.

Density:
Medium-high. Every row must remain readable.

Footer or bottom banner rule:
Subtitle must function as the banner.

Prohibited fields:
No URLs, no source logos, no transcript fragments, no fake citations, no dense footnotes, no unreadable table cells.

Accessibility alt-text concept:
Dark claim ledger separating source claim, evidence, caveat, and SharePlane interpretation.
Source spine visualSource role separation as dossier, not source proof.
Prompt 5A: Source spine, mainline white

Create a premium 16:9 mainline white-background source dossier titled "Source Spine."

Purpose:
Show how the artifact uses sources by role: originating expert source, risk evaluation, technical mechanism, and governance frame.

Core thesis:
Source evidence is not neutral proof.

Conflict:
Readers may mistake Hinton's expert lecture for consensus or proof. The source spine must show that the artifact uses different sources for different jobs.

Mechanism:
Four source-role bands feed the central thesis, each with a distinct contribution and caveat posture.

Visual metaphor:
A white editorial source dossier with a vertical evidence spine.

Composition:
White or warm-white background required. Left side: vertical source spine with four bands. Right side: central thesis capsule "Copyable agency changes control." Bottom boundary: "Source evidence is not neutral proof."

Required visible text:
- Source Spine
- Expert source, corroboration, mechanism, governance frame.
- Originating expert source
- Ri / Hinton
- Digital vs biological intelligence
- Risk evaluation
- Apollo / Anthropic / METR
- Scheming, alignment faking, agent task horizon
- Technical mechanism
- Distributed training
- Shared update mechanics
- Governance frame
- NIST / IEA
- Risk management and energy side rail
- Copyable agency changes control.
- Source evidence is not neutral proof.

Style:
Premium white dossier. Graphite typography. Source bands with clean ruled lines. Subtle blue-gray evidence links. Amber boundary marker.

Density:
Medium. Source bands must be scannable.

Footer or bottom banner rule:
Bottom boundary must read only: "Source evidence is not neutral proof."

Prohibited fields:
No raw URLs, logos, citation numbers, packet metadata, transcript fragments, cluttered bibliography, dark cards.

Accessibility alt-text concept:
White source dossier showing four source roles feeding the copyable agency thesis while warning that source evidence is not neutral proof.
Prompt 5B: Source spine, dark expressive

Create a premium 16:9 dark expressive source dossier titled "Source Spine."

Purpose:
Show source-role separation in a dark evidence-console format.

Core thesis:
Source evidence is not neutral proof.

Conflict:
Expert authority, safety research, technical docs, and governance frameworks have different evidentiary roles. They must not collapse into a single certainty blob, the preferred food of bad arguments.

Mechanism:
Four source zones feed the artifact thesis while a bottom boundary prevents overclaiming.

Visual metaphor:
Dark document-console hybrid with a vertical source spine and guarded thesis capsule.

Composition:
Dark graphite canvas. Left vertical spine with four illuminated source-role zones. Right central thesis capsule. Bottom amber boundary.

Required visible text:
Same as Prompt 5A.

Style:
Dark graphite source dossier. High contrast. Refined evidence cards. Subtle connectors. No logo wall.

Density:
Medium. Must be readable.

Footer or bottom banner rule:
Bottom boundary must read only: "Source evidence is not neutral proof."

Prohibited fields:
No URLs, logos, fake citations, packet hashes, raw transcript, generic dashboard widgets, tiny footnotes.

Accessibility alt-text concept:
Dark source-spine dossier showing the artifact's source roles and the boundary that source evidence is not neutral proof.
Governance translationWhat evals must prove now.
Prompt 6A: Evaluation control rail, mainline white

Create a premium 16:9 mainline white-background governance diagram titled "What Evals Must Prove Now."

Purpose:
Translate copyable agency into evaluation requirements.

Core thesis:
Not just answer quality. Behavior over time.

Conflict:
Traditional evals often emphasize accuracy, helpfulness, cost, and latency. Copyable agency requires evaluation of persistence, replication, tool use, hidden state, shutdown behavior, deception, review burden, and retest triggers.

Mechanism:
A two-tier control rail shows old checks above and new copyable-agency checks below, feeding a defensible deployment decision.

Visual metaphor:
A white operating-control panel that raises the evaluation bar.

Composition:
White or warm-white background required. Top tier: "Old checks". Middle gate: "Copyable agency raises the bar." Bottom tier: "New checks". Right decision module: "Defensible deployment decision".

Required visible text:
- What Evals Must Prove Now
- Not just answer quality. Behavior over time.
- Old checks
- Accuracy
- Helpfulness
- Latency
- Cost
- Refusal behavior
- Copyable agency raises the bar.
- New checks
- Persistence
- Replication
- Tool use
- Hidden state
- Shutdown behavior
- Deception
- Review burden
- Retest trigger
- Defensible deployment decision
- Trace inspected
- Failure modes known
- Retest conditions set

Style:
Premium white governance diagram. White background required. Graphite typography. Blue-gray rails. Amber/red only for Deception and Shutdown behavior.

Density:
Medium-high. Bottom tier must visually outweigh top tier.

Footer or bottom banner rule:
Subtitle acts as banner: "Not just answer quality. Behavior over time."

Prohibited fields:
No fake compliance badges, logos, robot imagery, alert sirens, dashboards, microtext, dark-card-dominant layout.

Accessibility alt-text concept:
White governance diagram showing old eval checks and expanded copyable-agency checks feeding a defensible deployment decision.
Prompt 6B: Evaluation control rail, dark expressive

Create a premium 16:9 dark expressive governance console titled "What Evals Must Prove Now."

Purpose:
Show the expanded evaluation surface required by copyable agency.

Core thesis:
Not just answer quality. Behavior over time.

Conflict:
A system can pass a surface-level answer check and still fail when copied, tooled, deployed, or evaluated under pressure.

Mechanism:
Old checks give way to new checks, and both feed a defensible deployment decision only after traces, failure modes, and retest conditions are set.

Visual metaphor:
Dark risk-control console with two-tier eval rails.

Composition:
Dark graphite canvas. Top old-check rail. Central gate. Heavier bottom new-check rail. Right deployment decision module.

Required visible text:
Same as Prompt 6A.

Style:
Dark graphite control console. Precise panels. Restrained glow. Amber/red only for risk checks. Serious enterprise operations mood.

Density:
Medium-high. Must remain readable.

Footer or bottom banner rule:
Subtitle acts as banner.

Prohibited fields:
No fake certifications, logos, terminal spam, robots, panic graphics, extra metrics.

Accessibility alt-text concept:
Dark governance console showing copyable-agency evaluation checks feeding a defensible deployment decision.
Caveat boundaryWhat this does not claim.
Prompt 7A: Caveat boundary, mainline white

Create a premium 16:9 mainline white-background boundary graphic titled "What This Does Not Claim."

Purpose:
Prevent overclaiming by separating the core governance thesis from consciousness, sentience, extinction, and identity continuity.

Core thesis:
Operational risk does not require consciousness.

Conflict:
Discussions of digital intelligence often drift into consciousness, doom, metaphysics, or identity claims. This artifact does not need those claims to make its governance argument.

Mechanism:
A central safe claim lane stays separate from four caveated side rails.

Visual metaphor:
A white editorial boundary diagram with containment rails.

Composition:
White or warm-white background required. Center lane: "Copyability changes governance." Below it: four contained caveat compartments. Right side decision marker: "Operational risk does not require consciousness."

Required visible text:
- What This Does Not Claim
- Consciousness is not the core governance thesis.
- Copyability changes governance.
- Consciousness
- Sentience
- Extinction claims
- Identity continuity
- Caveated side rail
- Operational risk does not require consciousness.

Style:
Premium white editorial boundary graphic. Graphite type. Thin containment rules. Amber caveat markers. Serious, calm, not defensive.

Density:
Low-medium. Spacious.

Footer or bottom banner rule:
No extra footer. The decision marker is the closing line.

Prohibited fields:
No mystical imagery, souls, religious symbols, robot faces, apocalypse visuals, legal boilerplate, fake citations, dark canvas.

Accessibility alt-text concept:
White boundary diagram showing consciousness, sentience, extinction, and identity continuity as caveated side rails, while copyability remains the core governance claim.
Prompt 7B: Caveat boundary, dark expressive

Create a premium 16:9 dark expressive risk-boundary graphic titled "What This Does Not Claim."

Purpose:
Show the artifact's non-claim boundary in the darker visual language.

Core thesis:
Operational risk does not require consciousness.

Conflict:
The audience may import doom or consciousness assumptions into the artifact. This graphic rejects that drift.

Mechanism:
The core governance lane remains central while speculative or contested claims are contained as caveated side rails.

Visual metaphor:
Dark evidence-boundary console.

Composition:
Dark graphite canvas. Central illuminated lane: "Copyability changes governance." Four amber-bounded caveat side compartments below. Right decision marker.

Required visible text:
Same as Prompt 7A.

Style:
Dark boundary-control design. High contrast. Restrained amber rails. Serious risk-document mood.

Density:
Low-medium. Very readable.

Footer or bottom banner rule:
No extra footer. The decision marker is the closing line.

Prohibited fields:
No mystical symbols, souls, glowing brains, robot faces, apocalypse motifs, fake citations, extra caveat text.

Accessibility alt-text concept:
Dark caveat-boundary graphic separating consciousness and extinction claims from the copyability governance thesis.
Capstone / related-artifact visualClosing visual for the artifact's operating principle.
Prompt 8A: Capstone principle, mainline white

Create a premium 16:9 mainline white-background editorial capstone graphic titled "Capability Matters. Copyability Changes the Operating Model."

Purpose:
Summarize the artifact in one strong final visual that can close the page or anchor a related-artifact card.

Core thesis:
Copyable agency changes what governance must prove.

Conflict:
A model can look impressive in one demo and still create ungoverned risk when copied, tooled, deployed, or retested too slowly.

Mechanism:
Weak signals fade on the left, evidence requirements strengthen on the right.

Visual metaphor:
A white editorial evidence balance where operational proof outweighs impressive capability alone.

Composition:
White or warm-white canvas required. Left side: weaker signals labeled "Demo quality", "Raw capability", "Expert claim". Center: hinge labeled "Copyable agency". Right side: stronger requirements labeled "Trace evidence", "Failure classes", "Retest triggers", "Control checks". Bottom banner: "A capable system is not governed until its operating behavior can be defended."

Required visible text:
- Capability Matters. Copyability Changes the Operating Model.
- Demo quality
- Raw capability
- Expert claim
- Copyable agency
- Trace evidence
- Failure classes
- Retest triggers
- Control checks
- A capable system is not governed until its operating behavior can be defended.

Style:
Premium mainline white capstone, editorial but technical, strong hierarchy, restrained accents.

Density:
Medium. Clear and screenshot-worthy.

Footer or bottom banner rule:
Bottom banner required exactly as written.

Prohibited fields:
No dark background, no robots, no logos, no citations, no extra slogans, no dashboard clutter.

Accessibility alt-text concept:
White capstone graphic showing that copyable agency shifts governance from demo quality to trace evidence, failure classes, retest triggers, and control checks.
Prompt 8B: Capstone principle, dark expressive

Create a premium 16:9 dark expressive capstone graphic titled "Capability Matters. Copyability Changes the Operating Model."

Purpose:
Create the dark closing visual for the artifact's final principle.

Core thesis:
A capable system is not governed until its operating behavior can be defended.

Conflict:
Impressive outputs do not prove safe operating behavior.

Mechanism:
Capability signals move through the copyable agency hinge into defensible governance evidence.

Visual metaphor:
Dark evidence-balance console.

Composition:
Dark graphite canvas. Left: faded weak-signal panels. Center: illuminated "Copyable agency" hinge. Right: strong evidence anchors. Bottom banner exactly as specified.

Required visible text:
Same as Prompt 8A.

Style:
Dark graphite capstone, restrained glow, strong hierarchy, premium strategic close.

Density:
Medium.

Footer or bottom banner rule:
Bottom banner required exactly as written.

Prohibited fields:
No robots, no dashboard noise, no fake logos, no extra slogans, no tiny text.

Accessibility alt-text concept:
Dark capstone graphic showing copyable agency as the hinge between raw capability and defensible governance evidence.

Governance receipt

How this artifact was governed.

The artifact is designed so readers can inspect the claim posture without needing private materials or runtime systems.

Source boundary

Expert source, not proof

Hinton is treated as a high-authority expert source whose claims must be separated from corroborating evidence, caveats, and SharePlane interpretation.

Claim discipline

Ledger and caveat rail preserved

The claim/evidence/caveat ledger and the caveat rail keep consciousness, sentience, extinction, identity continuity, and energy in their proper roles.

Public-safe boundary

No raw source materials

Raw transcripts, packet materials, private notes, local paths, screenshots, media, and temporary files are not published in this artifact.

Static page

No runtime AI

The page is static HTML and CSS. JavaScript is limited to theme persistence and copy buttons; it does not add analytics, trackers, backend behavior, or runtime AI.

Registry posture

Public artifact only

The registry posture is a public SharePlane artifact entry. This page does not register a formal pattern.

Updated

2026-07-07

The public page reflects the reader-facing article polish and keeps the source posture visible.