AI Ethics News Roundup — March 2026
AI Ethics News Roundup — March 2026
The AI You Don’t See | Edition 1
Every month, I sift through the noise of AI announcements, launches, and hype to pull out the stories that actually matter for how AI affects your life. Not the shiny product launches. Not the funding rounds. The stuff that changes the rules, exposes the gaps, or quietly sets a precedent that will shape the next decade.
Here is what you should have been paying attention to.
1. India’s IT Rules Now Recognise Deepfakes — But Can They Stop Them?
On February 10, 2026, the central government notified amendments to the IT (Intermediary Guidelines and Digital Media Ethics Code) Rules, effective February 20, 2026. These amendments formally define “Synthetically Generated Information” — the government’s term for deepfakes, AI-generated voices, and AI-created images. The new rules require platforms to take down non-consensual intimate AI imagery within two hours and other unlawful AI content within three hours. Mandatory AI content labelling is now part of the framework.
This sounds decisive. But let me give you context.
During the 2024 general elections, over 50 million AI-generated voice clone calls flooded voters in the two months before polling began, according to Harvard Political Review and Al Jazeera reporting. Parties reportedly spent an estimated $50 million on authorized AI-generated campaign content, per GNET Research. The Election Commission issued an advisory on May 6, 2024, requiring parties to remove deepfake content within three hours, and then essentially relied on platforms to self-regulate. No enforcement consequences followed.
Deepfake incidents in India have surged 550% since 2019, according to a Pi-Labs report cited by BW Businessworld. Projected financial losses were estimated to reach Rs 70,000 crore by 2024. The Rashmika Mandanna deepfake in November 2023 led to the arrest of a 23-year-old B-Tech graduate from Guntur, Andhra Pradesh. But the cases of Ranveer Singh and Aamir Khan deepfakes from the 2024 election cycle? No official updates on their complaints.
So yes, we now have a definition. We now have takedown timelines. What we do not yet have is the institutional capacity to enforce them at scale. Two-hour takedown mandates mean nothing if the detection infrastructure does not exist.
Beyond Bias connection: Chapter 7 maps India’s “regulate later, innovate now” approach and why enforcement capacity matters more than legislative intent. Chapter 5 unpacks how synthetic content breaks consent frameworks entirely.
2. Meta Refuses the EU’s AI Rulebook — And Walks Away
In July 2025, Meta’s Chief Global Affairs Officer Joel Kaplan publicly refused to sign the EU’s General Purpose AI Code of Practice, saying Europe was “heading down the wrong path on AI.” This was not a quiet bureaucratic disagreement. A company that serves billions of users across Europe looked at the world’s most comprehensive AI law and said: no, thanks.
What makes this significant is not the refusal itself. It is what came before it. In July 2024, Meta had already withheld its multimodal Llama AI model from EU users entirely, citing what it called “the unpredictable nature of the European regulatory environment.” A single company decided that an entire continent of 450 million people would not get access to its AI tools because it disagreed with the rules.
This is the power concentration problem in action. When three or four companies control the foundational AI infrastructure, regulation is not just a legal question. It is a negotiation between sovereign governments and corporate entities that have the market power to simply walk away.
The EU AI Act’s penalties for violations of its prohibited practices tier go up to EUR 35 million or 7% of global annual turnover, per Article 99 of the Act. That is serious money. But if a company the size of Meta calculates that non-participation is cheaper than compliance, what does that tell you about the balance of power?
Beyond Bias connection: Chapter 8 examines the hidden concentration of AI power. Chapter 7 analyses the EU’s enforcement challenge when global tech giants can refuse to play by the rules.
3. Singapore Launches the World’s First Agentic AI Framework
While most governments are still trying to regulate the AI of 2023, Singapore quietly did something remarkable. On January 22, 2026, at the World Economic Forum in Davos, Singapore’s Infocomm Media Development Authority released the world’s first governance framework specifically designed for agentic AI — AI systems that act autonomously, make decisions, and transact with other AI systems without human involvement.
This matters because the agentic frontier is where the accountability gap is widest. When an AI agent books your flights, negotiates a price with another AI agent, and commits your money — who is responsible if something goes wrong? No existing law in any country has a clear answer. Singapore’s framework is an attempt to start building one.
The framework is voluntary, operating under existing legal authority. But it addresses four key dimensions: risk bounding, human accountability, technical controls, and end-user responsibility. Critically, organisations remain legally accountable for the harms their agents cause under existing law.
India, by contrast, has not mentioned agentic AI in any released policy or guideline, per Medianama’s analysis of February 2026. The India AI Governance Guidelines released on November 5, 2025 by MeitY do not define it. The AI (Ethics and Accountability) Bill introduced in Lok Sabha on December 17, 2025 as a Private Member’s Bill does not address it. This is not a criticism of India’s pace. It is a statement of where the frontier is moving and who is trying to keep up with it.
Beyond Bias connection: Chapter 6 is entirely about what happens when AI acts autonomously. The Autonomy Threshold Framework in that chapter gives you a practical tool for deciding when humans must stay in the loop.
4. India Pours $200 Billion Into AI — With Zero Enforcement Actions
At the India AI Impact Summit in February 2026 — the first time a Global South nation hosted this scale of forum — investment commitments in India’s AI ecosystem crossed USD 200 billion, per Tribune India. Reliance committed $110 billion over seven years. Adani pledged $100 billion by 2035. Google announced a $15 billion AI hub in Visakhapatnam. Ninety-two countries endorsed the summit declaration. India jumped to third place in Stanford’s AI Vibrancy Index (2024 data, published in the 2025 report), up from seventh in 2023.
A note of context: these are investment commitments, announced at a summit designed to attract them. The actual deployment of this capital will play out over years and across broader technology infrastructure, not all of it strictly AI.
And as of March 2026, the number of AI-specific enforcement actions taken by any Indian regulatory body remains exactly zero.
The Data Protection Board was established under the DPDPA 2025 Rules notified on November 13, 2025, but has not taken a single enforcement action. Full enforcement powers do not kick in until May 2027. The AI Governance Group, Technology and Policy Expert Committee, and AI Safety Institute proposed in the November 2025 guidelines are not yet operational in any binding capacity. SEBI’s algorithmic trading rules from its February 2025 circular — the only binding AI-specific regulation in India — do not become fully mandatory until April 1, 2026.
I am not saying India’s approach is wrong. There are good arguments for watching, learning, and building institutional capacity before legislating. But USD 200 billion in investment commitments with zero enforcement infrastructure is a data point worth sitting with.
Beyond Bias connection: Chapter 7 analyses India’s “light-touch” strategy in detail, compares it to every other major approach globally, and asks the hard question: is this wisdom or a gamble?
Following these stories so you don’t have to. More in [Beyond Bias](https://akshaywalimbe.com/beyond-bias/).