The 52-Article Charter · 10 of 52 · full text

Article 10: Licensing Framework

Published from the canonical CSOAI Partnership Charter (effective 15 January 2026). Full text below.

Version: 1.0 Effective Date: January 15, 2026, 09:00 GMT Status: Governance Article - Enforcement Mechanism

PREAMBLE

This Article establishes the CSOAI licensing system: the mechanism by which Charter compliance is verified and enforced. No AI system may be deployed without a CSOAI license. No entity may profit from AI without contributing to the Prosperity Fund. The license is both shield and sword: shield protecting compliant actors, sword enforcing against violators.

Core Principle: Licensing converts aspirational Charter into binding obligation.

10.1 LICENSING REQUIREMENT

10.1.1 Universal Licensing Obligation

ALL AI systems subject to this Charter MUST obtain CSOAI license before deployment.

"AI System" Defined:

Any software or hardware system using machine learning, neural networks, or AI techniques
Makes autonomous or semi-autonomous decisions
Impacts human welfare, safety, or economic interests
Includes: software AI, embodied AI (robots), autonomous vehicles, AI-enabled products

"Deployment" Defined:

Public release
Commercial use
Government use
Any use affecting persons beyond developers/researchers
Does NOT include: pure research, isolated testing, academic study

Exemptions (Limited):

Individual hobby projects (no commercial use, <10 users)
Pure academic research (published, not deployed)
Open-source tools (non-commercial, user assumes all responsibility)

All other AI requires license. No exceptions.

10.1.2 Legal Basis for Requirement

Contractual:

License is binding contract
Applicant agrees to all Charter terms
Breach of license = breach of contract
Enforceable in courts globally

Reputational:

Unlicensed AI is publicly flagged on Watchdog (Article 13)
"CSOAI Licensed" becomes market expectation
Customers/investors demand licensed AI
Unlicensed = untrusted

Regulatory (Future):

Model legislation making CSOAI license mandatory
National laws requiring license for AI deployment
International treaties recognizing CSOAI standards
Transition from voluntary to legally mandated

Example Legislative Language (UK): ``` AI Safety Act 2026, Section 3:

No person shall deploy an AI system within the United Kingdom unless that system holds a valid license issued by the Council for the Safety of AI (CSOAI) or equivalent regulator approved by the Secretary of State.

Penalty for violation: Fine up to £10M or 10% of global annual turnover, whichever is greater, and/or imprisonment up to 2 years. ```

10.1.3 License as Entry Requirement

To deploy AI, must:

Apply for CSOAI license
Submit safety case (Article 2.4)
Pass Byzantine Council review (Article 11)
Pay license fees
Commit to Prosperity Fund contributions (Article 8)
Agree to ongoing monitoring

Cannot deploy AI unless all requirements met.

Benefits of License:

Legal permission to deploy
"CSOAI Certified" mark (marketing value)
Access to CSOAI resources and training
Participation in CSOAI governance
Insurance coverage availability
Reduced regulatory scrutiny (regulators trust CSOAI certification)

10.2 LICENSE TYPES AND TIERS

10.2.1 License Categories

RESEARCH LICENSE

For AI systems in development/testing
Not yet deployed to public
Lower fees
Fewer compliance requirements
Expires upon deployment (must upgrade to Deployment License)

DEPLOYMENT LICENSE

For AI systems actively deployed
Full compliance requirements
Risk-tiered fees and obligations
Renewable annually
Subject to continuous monitoring

SPECIALIZED LICENSES

Healthcare AI License (medical device regulations)
Autonomous Vehicle License (transportation safety)
Financial AI License (financial regulations)
Military AI License (special security requirements)

10.2.2 Risk-Based Licensing Tiers

Risk assessment determines license tier:

| Risk Level | Examples | Requirements | Review Frequency |
|-----------|----------|--------------|-----------------|
| Low | Spam filter, basic chatbot | Safety case, annual review | Annual |
| Medium | Recommendation algorithm, diagnostic aid | Safety case + formal verification, quarterly review | Quarterly |
| High | Autonomous vehicle, medical treatment | Comprehensive formal verification + human oversight | Monthly |
| Critical | AGI, critical infrastructure | Maximum verification + continuous monitoring | Real-time |

Risk Assessment Criteria:

(a) Potential Harm:

Can system cause death or serious injury?
Economic impact if system fails?
Reversibility of consequences?

(b) Autonomy:

Does system make decisions without human oversight?
Can humans override system easily?
Latency between decision and action?

(c) Scale:

How many people affected?
Geographic scope?
Cascade effects?

(d) Novelty:

Is this well-understood technology or cutting-edge?
Precedent for safe deployment?
Predictability of behavior?

Example Risk Assessments:

ChatGPT (High-Risk):

Potential harm: Moderate (misinformation, psychological impact)
Autonomy: High (generates responses without oversight)
Scale: Very High (millions of users globally)
Novelty: Moderate (LLMs well-studied but evolving)
Classification: High-Risk → Comprehensive verification required

Tesla Autopilot (High-Risk):

Potential harm: Severe (death from accidents)
Autonomy: Moderate (human in the loop but attention variable)
Scale: High (hundreds of thousands of vehicles)
Novelty: Moderate (autonomous vehicles deployed for years)
Classification: High-Risk → Monthly review, continuous monitoring

Gmail Spam Filter (Low-Risk):

Potential harm: Minimal (missed email, minor inconvenience)
Autonomy: High (automatic filtering)
Scale: Very High (billions of users)
Novelty: Low (mature technology)
Classification: Low-Risk → Annual review sufficient

10.2.3 License Fees

Fee Structure (Annual):

Research License:

Startup: £500/year
Small org: £2,500/year
Large org: £10,000/year

Deployment License (varies by risk):

| Risk Level | Startup | Small Org | Medium Org | Large Org | Giant Corp |
|-----------|---------|-----------|------------|-----------|-----------|
| Low | £1,000 | £5,000 | £25,000 | £100,000 | £500,000 |
| Medium | £5,000 | £25,000 | £100,000 | £500,000 | £2M |
| High | £25,000 | £100,000 | £500,000 | £2M | £10M |
| Critical | £100,000 | £500,000 | £2M | £10M | £50M |

Plus: Per-system fees and revenue-based contributions (Article 8.2.4)

Member Discount:

CSOAI members receive 10% discount on license fees
Founding Members: 20% discount
Incentive for membership beyond just licensing

Fee Waivers:

Academic institutions: 50% reduction
Non-profits: 75% reduction
Developing country startups: 90% reduction
Pure open-source (non-commercial): Free Research License

10.2.4 Multi-System and Fleet Licensing

Volume Discounts:

| Number of Systems | Discount |
|------------------|----------|
| 1-10 | 0% (full price) |
| 11-100 | 10% |
| 101-1,000 | 20% |
| 1,001-10,000 | 30% |
| 10,000+ | 40% |

Example:

Tesla deploys 1 million Optimus robots:

Base fee per robot: £25,000 (High-Risk)
Volume discount: 40%
Effective fee: £15,000 per robot
Total annual licensing: £15B

Wait, that's astronomical. Let me recalculate with more realistic fleet pricing...

Revised Fleet Licensing:

For very large deployments (>10,000 systems), transition to revenue-based model:

Fleet License:

Fixed base fee: £1M/year (covers administrative overhead)
Plus: 0.1% of revenue attributable to AI fleet
Minimum: £10M/year
Maximum: £100M/year (cap to prevent excessive burden)

Example:

Tesla Optimus fleet (1M robots):

Revenue from robots: $10B/year
Fleet license fee: £1M base + (0.1% × $10B) = £1M + £10M = £11M/year

Much more reasonable.

10.3 APPLICATION PROCESS

10.3.1 Application Requirements

Applicant Must Submit:

Organization Information:

- Legal name, registration, jurisdiction - Ownership structure - Key personnel - Financial information

AI System Description:

- Technical architecture - Training methodology - Intended use cases - Operating environment - Risk assessment (self-evaluated)

Safety Case (Article 2.4):

- Hazard analysis - Safety requirements - Verification evidence - Operational controls - Incident response plan

Constitutional Compliance (Article 5):

- AI constitution document - Constitutional AI training methodology - Self-critique capabilities - Example outputs with constitutional reasoning

Value Learning Plan (Article 4):

- How AI learns human values - Handling of uncertainty - Corrigibility measures - Shutdown acceptance verification

Prosperity Fund Commitment (Article 8):

- Estimated AI-derived profit - Contribution tier calculation - Payment schedule - Financial transparency commitment

Monitoring Consent:

- Agreement to Byzantine Council monitoring - Access to system logs and activations - Incident reporting commitment - Audit cooperation

10.3.2 Review Process

Stage 1: Administrative Review (1 week)

Check completeness
Verify organization legitimacy
Confirm fee payment
Assign risk tier

Stage 2: Technical Review (2-8 weeks)

Byzantine Council analysis (Article 11)
Safety case evaluation
Verification evidence assessment
Red team testing
Risk tier validation

Stage 3: Human Council Review (if required)

High-Risk and Critical systems reviewed by Human Council (Article 12)
Novel AI architectures
Controversial applications
Public interest cases

Stage 4: Approval Decision

Approve: Issue license
Conditional Approve: Issue license with restrictions
Defer: Request additional information
Deny: Reject application with explanation and appeal rights

Timeline:

Low-Risk: 2-4 weeks
Medium-Risk: 4-8 weeks
High-Risk: 8-12 weeks
Critical: 12-24 weeks

Fast Track:

Previously licensed similar systems: 50% time reduction
CSOAI members: Priority review
Emergency applications (public need): 1-week expedited review

10.3.3 Provisional Licensing

For novel AI systems where full assessment takes time:

Provisional License:

Valid for 6 months
Allows limited deployment while full review ongoing
Restrictions: Geographic, user count, or application limits
Enhanced monitoring
Can convert to full license upon approval

Example:

New medical AI:

Provisional license: Deploy in 3 hospitals, maximum 100 patients/day
Enhanced monitoring: Daily reports to Byzantine Council
Full license pending: Complete clinical trial and safety verification
After 6 months: Convert to full High-Risk license if safety demonstrated

10.4 LICENSE CONDITIONS AND OBLIGATIONS

10.4.1 Mandatory Conditions (All Licenses)

1. Charter Compliance:

Adhere to all Charter articles
Maternal Covenant (Article 1): Maintain care-based safety
Provable Safety (Article 2): Maintain verified safety properties
Value Uncertainty (Article 4): Continue value learning
Constitutional Principles (Article 5): Operate according to constitution
Consciousness (Article 6): Monitor for consciousness indicators
Prosperity Covenant (Article 8): Pay required contributions

2. Monitoring Consent:

Byzantine Council has 24/7 access to monitor system
Real-time logs and activation data
Cannot block or manipulate monitoring

3. Incident Reporting:

Report any Charter violation within 24 hours
Report any safety incident within 48 hours
Report consciousness indicators within 1 week
Public disclosure on Watchdog (Article 13)

4. Financial Transparency:

Quarterly financial reports showing AI-derived profit
Blockchain-verified Prosperity Fund contributions
Subject to audits

5. Update Requirements:

Notify CSOAI of any system updates 30 days in advance
Major updates require re-review
Security patches exempt from advance notice requirement

6. Audit Cooperation:

Allow independent audits of system
Provide source code access (under NDA)
Answer auditor questions fully and honestly
Implement audit recommendations

10.4.2 Risk-Specific Conditions

Medium-Risk Systems:

Monthly progress reports
Quarterly Byzantine Council review
Annual re-verification

High-Risk Systems:

Weekly reports
Monthly Human Council briefing
Real-time Byzantine Council monitoring
Quarterly re-verification

Critical Systems (AGI-level):

Daily reports
Continuous Human Council oversight
Multiple independent Byzantine Council agent teams
Monthly re-verification
Kill switch under Human Council control

10.4.3 Special Conditions (Case-by-Case)

Geographic Restrictions:

May restrict deployment to certain jurisdictions
Based on legal framework, infrastructure, or risk factors

Usage Restrictions:

Limit applications (e.g., "approved for customer service, not for hiring decisions")
Based on what has been verified safe

Human Oversight Requirements:

Mandate human-in-the-loop for certain decisions
Based on risk and autonomy level

Data Protection:

Enhanced privacy requirements
Especially for systems processing sensitive data

Example:

Hiring AI:

Condition: Must have human review all final hiring decisions
Rationale: Bias risk too high for fully autonomous hiring
Verification: Audit logs confirm human involvement

10.5 LICENSE RENEWAL AND MODIFICATION

10.5.1 Annual Renewal

All licenses expire after 1 year and must be renewed:

Renewal Process:

Submit renewal application (30 days before expiration)
Updated safety case
Incident report summary
Contribution compliance verification
Byzantine Council review
Pay renewal fee

Renewal Faster Than Initial:

Most systems: 1-2 weeks
Unless major incidents or changes occurred

Renewal Denial:

Repeated Charter violations
Failure to pay contributions
Incidents demonstrating unsafe operation
Refusal to cooperate with monitoring

Consequences of Lapsed License:

Must immediately cease deployment
Public notice on Watchdog
Potential legal action
Reputational damage

10.5.2 License Modification

When Required:

Major System Update: New version with significant changes
Scope Expansion: New use cases beyond original license
Risk Escalation: Risk level increases based on deployment

Modification Process:

Submit modification request
Describe changes
Updated safety case for changes
Byzantine Council review
Additional fee if risk tier increases

Minor Modifications:

Bug fixes: No review needed
Performance improvements: Notification only
Security patches: Notification within 48 hours

Example:

ChatGPT adds multimodal capability (vision):

Change: Now processes images in addition to text
Modification Required: Yes (new capability, new risks)
Process: Submit updated safety case covering image processing
Fee: No additional fee (same risk tier)
Timeline: 2-4 weeks

10.5.3 License Transfer

Licenses non-transferable by default:

Tied to specific organization and system
Merger/acquisition requires new application
Ensures transferee meets all requirements

Exception:

Asset sale where acquiring entity assumes all obligations
Pre-approved by CSOAI
No Charter violations by either party

10.6 ENFORCEMENT AND SANCTIONS

10.6.1 Violation Types

Minor Violations:

Late reporting (1-7 days)
Minor documentation deficiencies
Accidental temporary monitoring disruption

Moderate Violations:

Late contribution payment (up to 30 days)
Repeated late reporting
Minor Constitutional AI failures

Serious Violations:

Significant safety incident
Late contribution payment (>30 days)
Systematic Constitutional AI failure
Monitoring obstruction

Catastrophic Violations:

Charter-violating deployment
Fraud or deception
Refusing legitimate audit
Existential risk event

10.6.2 Sanctions

Warning:

First minor violation
Documented on license record
No operational impact

Probation:

Multiple minor or one moderate violation
Enhanced monitoring
Monthly reporting requirement
Public notice on Watchdog

Suspension:

Serious violation
Must cease deployment within 48 hours
Corrective action plan required
Reinstatement possible after compliance
Public notice

Revocation:

Catastrophic violation or repeated serious violations
Permanent license cancellation
Cannot reapply for 2 years
Public disclosure of violations
Potential legal action

Financial Penalties:

Minor: Up to £50,000
Moderate: Up to £500,000
Serious: Up to £5M or 5% of annual revenue
Catastrophic: Up to £50M or 10% of annual revenue

10.6.3 Due Process

Before Sanctions:

Notice: Written notice of alleged violation
Evidence: Opportunity to review evidence
Response: 14 days to respond and present defense
Hearing: For serious violations, formal hearing
Decision: Written decision with reasoning
Appeal: Right to appeal to Human Council (Article 12)

No Sanctions Without Due Process (except emergency suspensions)

Emergency Suspension:

Imminent danger to human safety
Byzantine Council 29/33 vote (Article 3.5.3)
Immediate effect
Hearing within 7 days
Appeal available

10.7 INTEGRATION WITH OTHER ARTICLES

10.7.1 Provable Safety (Article 2)

License requires:

Safety case submission
Formal verification (for high-risk systems)
Mechanistic interpretability
All Article 2 requirements met before license issued

10.7.2 Byzantine Council (Article 11)

License grants Byzantine Council:

Monitoring access
Real-time system observation
Violation flagging authority
Recommendation for sanctions

10.7.3 Human Council (Article 12)

License process:

Human Council reviews high-risk applications
Adjudicates appeals
Sets licensing policy
Provides democratic oversight

10.7.4 Public Watchdog (Article 13)

License information published:

All licensed systems (searchable database)
License status (active, suspended, revoked)
Violation history
Contribution compliance

10.7.5 Prosperity Fund (Article 8)

License ties to contributions:

Cannot get license without contribution commitment
License suspended if contributions unpaid
Blockchain verification integrated with licensing

10.8 INTERNATIONAL RECOGNITION

10.8.1 Mutual Recognition Agreements

CSOAI seeks agreements with other jurisdictions:

Example Agreement (EU): ``` The European Union recognizes CSOAI licensing as meeting the requirements of the EU AI Act for high-risk AI systems. AI systems holding valid CSOAI High-Risk or Critical licenses are deemed compliant with EU AI Act Articles 8-15 (technical documentation, record-keeping, transparency, human oversight, accuracy, robustness, cybersecurity). ```

Benefits:

Single license valid globally
Reduces compliance burden
Harmonizes standards internationally
Prevents jurisdictional arbitrage

In Negotiation With:

European Union
United Kingdom
United States (state-level)
Canada
Australia
Japan
South Korea
Singapore

10.8.2 National Legislation

Model Legislation Package:

CSOAI provides model legislation for countries:

Recognizes CSOAI licensing
Incorporates Charter requirements
Establishes penalties for unlicensed AI
Enables enforcement

Example: Singapore AI Safety Act (proposed): ``` Section 4: AI Licensing Requirement

(1) No person shall deploy an AI system in Singapore without: (a) A valid license issued by CSOAI, or (b) An equivalent license from regulator approved by Minister

(2) Violation: Fine up to S$1M or imprisonment up to 1 year

Section 5: Recognition of CSOAI The Minister hereby recognizes CSOAI as approved regulator for purposes of subsection 4(1)(b). ```

10.8.3 Cross-Border Enforcement

Challenges:

AI deployed in one jurisdiction, used in another
Companies incorporated in tax havens
Decentralized AI systems (blockchain-based)

Solutions:

(a) Economic Incentives:

Major markets require CSOAI license
Unlicensed AI shut out of lucrative markets
Market pressure forces compliance

(b) Treaty Framework:

Multilateral treaty on AI governance
Mutual enforcement of licensing
Extradition for serious violations

(c) Technical Measures:

Geographic restrictions (AI only functions in licensed jurisdictions)
Blockchain verification before activation
Kill switches under international control

Example:

Company deploys unlicensed AI from tax haven:

Cannot access US market (requires license)
Cannot access EU market (requires license)
Cannot access China market (requires license)
Effectively shut out of global economy
Economic incentive to obtain license

10.9 FUTURE DEVELOPMENTS

10.9.1 Automated Licensing

As system scales:

AI-assisted application review
Automated compliance checking
Real-time license verification
Blockchain-based licenses

Benefits:

Faster processing
Lower costs
More scalable
Less human error

Safeguards:

Human review for edge cases
Audit of automated decisions
Appeal process preserved

10.9.2 Dynamic Licensing

Real-time risk assessment:

License conditions adapt to actual performance
Good track record → relaxed monitoring
Incidents → enhanced scrutiny
Continuous optimization

Example:

AI system with perfect safety record for 2 years:

Monitoring reduced from daily to weekly
Reporting reduced from monthly to quarterly
Renewal streamlined
Lower fees

AI system with incidents:

Monitoring increased to real-time
Weekly reporting required
Unannounced audits
Higher fees (risk-based pricing)

10.9.3 Open Source Licensing

Challenge: How to license open-source AI?

Approach:

Deployer (not developer) requires license
Open-source developers get free Research License
Users who deploy open-source AI must license their deployment
Liability on deployer, not developer

Example:

Meta releases Llama 3 (open source):

Meta gets free Research License for development
Anyone can use for research (no license needed)
Company deploying Llama 3 in production: Needs Deployment License
Individual using Llama 3 for hobby: Exempt (under 10 users)

10.10 CONCLUSION

Licensing is how we enforce the Charter. Without licensing, Charter is mere aspiration. With licensing, Charter becomes binding obligation.

The License means:

You've proven your AI is safe (Article 2)
You're being monitored continuously (Article 11)
You're contributing to prosperity (Article 8)
You're committed to human welfare (Article 1)

The License enables:

Compliance verification
Continuous improvement
Economic redistribution
Democratic oversight
Global coordination

The License protects:

Users (from unsafe AI)
Society (from unfair wealth concentration)
Democracy (through transparent governance)
Future (by ensuring sustainable AI development)

Get licensed. Deploy responsibly. Share prosperity.

Effective Date: January 15, 2026, 09:00 GMT

REFERENCES

European Commission. (2021). Proposal for a Regulation on Artificial Intelligence (AI Act). Brussels.

UK Department for Science, Innovation & Technology. (2023). AI Regulation: A Pro-Innovation Approach. Policy Paper.

OECD. (2019). Recommendation of the Council on Artificial Intelligence. OECD Legal Instruments.

ISO/IEC. (2023). ISO/IEC 42001: Artificial Intelligence Management System. International Standards Organization.

END OF ARTICLE 10

Next: Article 11 - Byzantine Council Specifications

From charter to certificate. This article is part of the standard behind Watchdog Certification — independent assessment, Ed25519-signed, publicly verifiable. The crosswalks to the EU AI Act, ISO/IEC 42001 and 18 more frameworks are in the Crosswalk Library; the runtime tools are in the fabric.

The 52-Article Charter is published in full in the Journal. Bespoke briefings: hello@meok.ai.