AI Governance vs National Security: Pentagon–Anthropic Dispute & Ethical Risks

The U.S. Pentagon–Anthropic dispute over removal of AI safeguards highlights tensions between national security imperatives and ethical constraints in AI deployment.

Background

• Government Pressure vs Ethical Limits: The US government reportedly asked Anthropic to remove safeguards against mass surveillance and autonomous weapons, which the company refused.
• Shift to Alternative Providers: The government’s move to partner with OpenAI raises concerns about who governs AI when states bypass ethical resistance.
• Core Issue: When governments become primary users, the risk of AI being used for harmful purposes increases without independent checks.

Appropriate vs Dangerous Uses of AI

• Appropriate Use (Well-defined, Controlled Contexts): AI should be deployed in areas with clear objectives, limited scope, and measurable outcomes, ensuring human oversight and public benefit.
• Dangerous Use (Open-ended, Unchecked Power): Facial recognition, mass surveillance, and autonomous weapons concentrate excessive state power, undermine civil liberties, and reduce meaningful human control over critical decisions.
• Normative Principle for Governance: Adopt a “do no harm” approach by strictly prohibiting autonomous lethal systems and tightly regulating high-risk AI applications to safeguard rights and democratic freedoms.

Structural Risks- The Three Governance Traps

• Efficiency Trap (Labour Substitution): AI-driven automation often replaces human labour in public services without clear evidence of efficiency gains or improved outcomes, risking job losses and weakened service delivery.
• Function Creep: Data collected for specific welfare purposes (such as ration distribution) is gradually repurposed for surveillance or policing, often without transparency or informed public approval.
• Illusion of Consent: Low digital literacy leads citizens to mechanically accept terms and conditions, resulting in uninformed consent for extensive data collection and use.

Data, Privacy and Political Economy of AI

• Privacy as a Fundamental Right: The K.S. Puttaswamy (2017) judgment recognises privacy as intrinsic to dignity, implying that personal data cannot be treated merely as an economic resource.
• Myth of ‘More Data equals Better AI’: Advances in efficient models (e.g., DeepSeek) demonstrate that high-quality algorithms can deliver strong outcomes without relying on excessive data extraction.
• Risk of Data Monetisation: Framing citizen data as a “national asset” risks commodifying individual rights and legitimising intrusive data collection practices by the state and private actors.

Enroll now for UPSC Online Course

Breakdown of Regulatory Logic and Accountability

• Violation of the Golden Rule: Unlike sectors such as mining, where regulation precedes operation, AI is often deployed first and regulated later, reversing the standard governance sequence.
• Case Study (Jharkhand): AI-based fingerprint authentication failures led to the denial of rations to beneficiaries, yet no accountability was fixed on the private company responsible.
• Governance Gap: Absence of clear liability and accountability frameworks allows harms caused by AI systems to persist without effective redress.

Illusion of Voluntariness in Digital Governance

• De facto Mandatory Systems: Services like Digi Yatra are framed as voluntary but, in practice, impose systemic disadvantages on those who opt out, making participation almost unavoidable.
• Coercive Choice Architecture:  Consent loses its meaning when alternatives are inefficient, time-consuming, or exclusionary, effectively nudging citizens into accepting digital systems.

Strategic Autonomy and the Dependency Trap

• Big Tech Pressure Narrative: The fear of falling behind countries like the US and China is used to push governments toward rapid AI adoption and increased funding, often without adequate safeguards.
• Need for Indigenous Capability: India should prioritise foundational research and capacity building, following the ISRO and nuclear programme approach, rather than relying heavily on foreign AI systems.
• Risk to Technological Sovereignty: Dependence on external AI models can erode strategic autonomy, weaken domestic innovation ecosystems, and limit long-term technological self-reliance.

Way Forward

• Use AI for Clearly Defined Public Good Problems:  Deploy AI in areas with clear objectives, limited scope, and measurable outcomes to ensure effectiveness and minimise unintended harms.
• Prohibit High-risk Applications: Ban or strictly regulate uses such as mass surveillance and autonomous weapons that threaten civil liberties and human control.
• Regulation Before Deployment: Establish robust legal, ethical, and accountability frameworks prior to large-scale implementation of AI systems.
• Ensure Accountability and Transparency: Fix clear liability on both government and private actors and ensure transparency in AI decision-making to enable redressal.
• Strengthen Domestic Ecosystem:  Invest in basic science, research, and indigenous AI development to build long-term technological self-reliance and avoid external dependency.

Check Out UPSC CSE Books

Visit PW Store

Conclusion

Governments must adopt a restrained and principle-based approach to AI, leveraging its benefits for public welfare while safeguarding fundamental rights, ensuring accountability, and preventing the erosion of democratic freedoms.

DOWNLOAD PDF