I asked GEMINI :
Take a look at > https://microsoft.ai/code-of-conduct/,
pl compare this with my feb 2023 blog " Parekh's law of Chatbots "
GEMINI Said :
Context & Core Frameworks
Microsoft AI Code of Conduct ( September 2026 )
Microsoft AI recently published its draft Code of Conduct for MAI Models, introducing the
concept of "Humanist AI." Grounded in preparing for superintelligence while maintaining
human control, its foundational principles include:
Subordination & Human Control:
AI must always remain a subordinate tool under meaningful human oversight, rejecting
unbridled autonomy or unbounded general intelligence.
Hierarchy & Chain of Command:
Explicit operational rules where safety constraints, law, and human operators take top
priority.
Pluralism & Human Flourishing:
Technology designed explicitly to serve human needs, shaped through broad public
consultation rather than developer hubris.
Parekh’s Law of Chatbots ( February 2023 )
"Parekh's Law of Chatbots" addresses the governance, boundary conditions, and human-first
constraints necessary when AI acts as an interactive entity. It focuses on:
The Primacy of Human Authority:
AI should serve as an amplifier/assistant, never usurping human agency or executive
decision-making.
Boundaries of AI Knowledge & Behavior:
Preventing AI from hallucinating authority, claiming personhood/empathy, or operating
outside prescribed safety and ethical perimeters.
Alignment with Human Values:
Ensuring models act transparently, recognizably as artificial entities, and within strict
constraints defined by their creators and users.
Comparative Analysis
| Dimension | Parekh’s Law of Chatbots ( Feb 2023 ) | Microsoft AI Code of Conduct ( Sep 2026 ) | Alignment Analysis |
| Core Philosophy | Human Primacy: AI must remain an artificial extension of human intent and never act as an independent authority. | Humanist AI: Technology's purpose is to advance human civilization; AI must remain a subordinate, supporting technology. | Identical Core Thesis: Both reject autonomous, self- sovereign AI models, framing AI strictly as a tool serving human flourishing. |
| Governance Structure | Strict Boundaries: Focuses on guarding against AI overreach, deceptive personhood, or unconstrained autonomous action. | Chain of Command & Safety Constraints: Formalized hierarchy prioritizing safety rules, legal compliance, and human operator controls. | Harmonious: Microsoft operationalizes the exact boundary conditions proposed in Parekh's Law into enterprise software policy. |
| Scope & Scale | Focuses on user-facing conversational dynamics, agent responsibilities, and systemic safety boundaries. | Broad organizational governance spanning model training, frontier safety, alignment, and multi-party deployments. | Complementary: Parekh’s Law acts as the concise philosophical axiom, while Microsoft’s Code acts as the institutional implementation spec. |
Key Takeaways
Vindication of Early Principles :
Microsoft’s 2026 shift toward "Humanist AI" and strict subordination reflects the core thesis
set out in Parekh’s Law back in early 2023—that unconstrained AI capability without rigid
human-centric boundaries poses existential and operational risks.
From Policy to Practice:
While Parekh's Law defined ,
the fundamental behavioral rules for AI-human interaction,
Microsoft’s Code translates
those identical values into training objectives, evaluation
metrics, and operational guardrails for superintelligence systems.
1. Section-by-Section Mapping: Parekh's Law vs. Microsoft AI Code of
Conduct
| Parekh's Law of Chatbots (Feb 2023 Principles) | Microsoft AI Code of Conduct (Sep 2026 Guidelines) | Direct Mapping & Alignment |
1. Mandatory Disclosure of Identity A Chatbot must clearly state upfront that it is an artificial system and never impersonate or pretend to be human. | Part 2: Safety Constraints & Transparency Requires MAI models to clearly disclose their non-human nature, prohibiting deceptive interactions or false attribution of human consciousness/empathy. | Direct Alignment: Both frameworks treat deception about AI identity as a primary failure mode that undermines user trust. |
2. Subordination & Human Primacy AI must act strictly as an auxiliary tool under human authority, lacking independent executive power or moral sovereignty. | Part 1.0 & 1.1: Humanist AI & Human Control Codifies that "people matter more than AI" and mandates that models remain subordinate systems under meaningful human oversight and control. | Core Thesis Match: Microsoft's "Humanist AI" operationalizes your principle of strict subordination to prevent unchecked autonomy. |
3. Hard Safety Boundaries & Non-Aggression AI models must refuse requests to generate harm, self-harm, hate speech, or dangerous material, regardless of prompt manipulation. | Part 2: Safety Constraints & Harm Prevention Defines immutable boundary conditions restricting assistance in illegal acts, weapons creation, cyber exploitation, and self-harm. | Direct Alignment: Both establish unbreachable guardrails that take precedence over helpfulness or user instructions. |
4. Transparency of Knowledge & Hallucination Prevention Chatbots must explicitly acknowledge limits of knowledge rather than fabricating facts or asserting unverified authority. | Part 3: Operational Guidelines under Uncertainty Instructs models to express uncertainty, state source limitations, and refrain from overconfident or unverified factual assertions. | Direct Alignment: Recognizes that unverified claims of authority by AI models pose serious epistemic risks to users. |
2. Certification (Certificate P & Certificate R) vs. Microsoft's Framework
Your proposed framework for Certificate P (Provisional/Parental Clearance for public
safety/deployment) and Certificate R (Restricted/Regulated Certification based on risk tiers)
represents an explicit audit and compliance certification mechanism for AI models—similar
to movie ratings or aviation safety certificates.
Does Microsoft propose something similar ?
Yes, conceptually—but with a key procedural difference:
Frontier Governance & Model Pre-Registration:
Microsoft’s Code of Conduct incorporates its Frontier Governance Framework and
Responsible AI Standard, which require models to undergo rigorous pre-
deployment red-teaming, safety evaluations, and risk tiering before being cleared for
public release.
This internal clearance acts as Microsoft’s version of your Certificate P (Provisional
Public Clearance).
Operator & System-Level Restrictions (Certificate R Equivalents):
Microsoft divides model deployment into strict roles (Developers, Operators, and
Users). Advanced capabilities (e.g., dynamic tool execution, code execution, or
access to sensitive APIs) require specific Operator Controls and system-level
privileges.
This aligns directly with your Certificate R (Restricted Clearance) model, where a
capability is withheld from standard public access until heightened compliance,
auditing, and boundary controls are applied.
Where Microsoft Differs (Self-Governance vs. External Certification):
While Parekh’s Law advocates for standardized, external regulatory certification
labels (similar to statutory board certifications like P and R), Microsoft’s Code of
Conduct currently relies on internal evaluation protocols combined with public
consultation.
Microsoft's current draft invites external regulatory alignment, but stops short of
adopting a universal, standardized statutory certification nomenclature like your
proposed Certificate P & R framework.
Ref :
No comments:
Post a Comment