India AI Impact Summit 2026

artificial-general-intelligence-a-new-paradigm-of-safety-security-privacy-ethics-and-governance

6 min read

TRACK:Safe and Trusted AI
SPEAKERS
  • Mr. Atul Kumar, Government Initiatives & Global Trade . DSCI
  • Mr. Avneesh Pandey, SEBI
  • Mr. Hendrikus G.J. (Harry) Verweij, Netherland
  • Mr. Kenny Kesar, Wipro
  • Mr. Simonas, INFOBALT
  • Mr. Vinayak Godse, Data Security Council of India
  • Ms. Alexandra Bech Gjørv, SINTEF
  • Ms. Aradhana Gupta, Data Security Council of India
  • Ms. Nicole Foster, Amazon
SESSION OVERVIEW

Artificial general intelligence will intensify existing challenges around safety, security, privacy, ethics, and governance. This session will examine how current trajectories in advanced AI raise near-term governance questions, particularly for developing economies. It will explore institutional readiness, policy gaps, and risk pathways across data, compute, markets, and deployment contexts, and will surface practical considerations for anticipatory governance that can inform responsible decision-making before irreversible dependencies take hold.

VIDEO RECORDING

Artificial General Intelligence: A New Paradigm of Safety, Security, Privacy, Ethics, and Governance

Abstract

The panel examined the accelerating trajectory toward artificial general intelligence (AGI) and its ramifications for safety, security, privacy, ethics, and governance, especially for developing economies such as India. Participants debated definitions and timelines for AGI, the hardware and data requirements, the risks of over‑reliance on AI for critical thinking, and the need for anticipatory governance frameworks that span technical, regulatory, and societal dimensions. Concrete ideas—ranging from “anchor‑control” mechanisms and robust rollback strategies to education of policymakers and the public—were explored as early‑stage safeguards against the emerging risk pathways.

Detailed Summary

1. Opening Remarks – Framing the Topic

  • Kenny Kesar opened the session by noting the rapid acceleration of AI capabilities since 2020, culminating in the “powerful AI” launches of 2023‑24.
  • He described the panel’s purpose: to demystify Artificial General Intelligence (AGI), to gauge its timeline, and to explore the security, privacy, ethical and governance challenges that AGI will amplify.

2. Defining AGI – Divergent Views

2.1 A Pragmatic Definition

  • Kenny offered a common industry shorthand: AGI must reason, learn, adapt, transfer knowledge, and operate beyond narrow task boundaries.
  • He cautioned that the field still lacks a precise timetable for when true AGI will appear.

2.2 Human‑Level Task Performance (Speaker “Nir” – possibly Vinayak Godse)

  • Proposed a simplistic baseline: “an AI that can perform every human task with professional‑level accuracy.”
  • Acknowledged the difficulty of measuring “every task” and noted that public perception is already blurring the line—‑​e.g., 50 % of Israelis trust generative AI more than a friend.
  • Predicted a 3‑7 year horizon before that baseline is reached, based on user trust trends rather than pure technical metrics.

2.3 Research‑Centric View (Alexandra Bech Gjørv)

  • Stressed that hardware innovations (low‑latency, energy‑efficient neuromorphic chips, edge computing) are prerequisites for real‑time, context‑aware reasoning.
  • Highlighted hierarchical reflex‑reasoning systems and embodied multimodal learning being pursued at SINTEF.
  • Raised a crucial privacy bottleneck: achieving human‑level situational awareness requires massive personal data, which collides with data‑protection regulations.

3. Technical Enablers & the Role of Compute

3.1 Accuracy as a Trajectory Toward Generality

  • Kenny linked accuracy (“five‑nines”) to the path toward AGI: each additional “9” demands roughly an extra 1‑2 years of research and larger models.
  • He argued that research‑driven AI (self‑generation of ideas) is the next frontier; once AI can research like a human brain, the system approaches general intelligence.

3.2 Compute – The “Fuel” of the Race

  • Kenny described a current investment frenzy in compute infrastructure (GPUs, TPUs, specialized ASICs).
  • He warned of a potential bubble, noting that efficiency improvements (e.g., sparsity, model‑distillation) could curb over‑investment.
  • Alexandra added that compute is only one link in a chain; energy, data, algorithms, and human talent are equally decisive.

3.3 Metaphor: Preparing for an Unknown Transportation Revolution

  • Alexandra likened today’s scramble to a 19th‑century prophecy about a new transport mode: some built airports, others railways.
  • The point: over‑focusing on a single technology (compute) ignores complementary needs—‑​energy supply, data pipelines, skilled workforce, and regulatory frameworks.

4. Governance, Policy, and the Human Factor

4.1 Government Readiness & Education

  • Mr. Atul Kumar stressed that governments must avoid “compute‑only” strategies and invest in critical‑thinking education and skill‑building for policymakers.
  • He cited the “human oversight” paradox: while humans are needed to set ethical boundaries, they are prone to bias and moral luck in high‑stakes AI decisions (e.g., autonomous vehicle crash dilemmas).

4.2 Regulation Across Levels of Risk

Risk LevelDescription (from panel)Illustrative Points
Classical Technical RisksPrivacy, security, fraud, cyber‑threats.AGI could emulate a CEO and make high‑impact decisions; current cyber‑defence must stay a step ahead.
Human‑Health & Mental‑Health RisksAI‑generated content affecting wellbeing.Over‑reliance on AI may erode critical‑thinking muscles, leading to cognitive atrophy.
Social‑Cohesion RisksEmpathy loss, bullying, addiction.AI‑driven recommendation loops can amplify echo chambers and reduce real‑world interaction.
Macro‑Societal RisksThreats to democracy, geopolitical manipulation.AI‑generated “information universes” can weaponise public opinion; example: deep‑fake propaganda in the Ukraine‑Russia conflict.

4.3 International Coordination

  • Mr. Avneesh Pandey (SEBI) and Mr. Hendrikus Verweij highlighted the need for cross‑border standards to prevent fragmented regulation that could be gamed by multinational AI firms.
  • Ms. Nicole Foster (Amazon) underscored the industry’s role in publishing transparent model cards, watermarking, and auditable logs to aid regulators.

5. Societal Implications – Critical Thinking & Media Literacy

5.1 Diminishing Critical Thinking

  • Simon argued that AI presently outsources critical‑thinking: users ask the model for nuanced analyses, and the model’s output becomes the user’s “thinking”.
  • This creates a feedback loop where AI learns from AI‑generated content, eroding the diversity of intellectual sources.

5.2 Education as a Countermeasure

  • Simon advocated a bottom‑up approach: embed media‑literacy, critical‑analysis, and data‑privacy awareness in school curricula and public outreach.
  • He also called for industry‑driven “AI operating procedures” (AOPs) akin to SOPs, to embed bias‑checks and ethical guardrails directly into model deployment pipelines.

6. “Anchor‑Control” Concepts – Early‑Stage Safeguards

  • The panel collectively identified technical “anchor” mechanisms that could be deployed before AGI arrives:

    1. Watermarking & provenance labeling of AI‑generated content.
    2. Robust rollback & resilience planning (e.g., “operate without electricity” drills).
    3. Sector‑wide AI Operating Procedures (AOPs) for bias and safety verification.
    4. Multi‑stakeholder governance bodies that combine regulators, academia, and industry to co‑design moral frameworks.

  • Mr. Atul Kumar highlighted the necessity of training politicians to understand the trade‑offs between “human moral luck” and algorithmic determinism.

7. Closing Remarks & Announcements

  • The moderator thanked participants, announced a photo‑shoot with the panel, and introduced a launch of the “AI Cybersecurity Terminal” – a new initiative aimed at providing real‑time threat intelligence for AI‑driven attacks.

Key Takeaways

  • AGI Timeline – Consensus among panelists leans toward a 3‑7 year horizon before AI can match professional‑level human task performance, driven more by trust adoption than pure hardware breakthroughs.
  • Definition of AGI – Agreed that AGI must be general, adaptable, and non‑narrow; it should integrate reasoning, learning, knowledge transfer, and contextual awareness.
  • Compute Is Necessary but Not Sufficient – Massive compute fuels progress, yet energy, data, hardware design, and human talent are equally critical; over‑investment in compute alone risks a bubble.
  • Privacy Bottleneck – Human‑level situational awareness demands large volumes of personal data, creating tension with privacy regulations and ethical constraints.
  • Multi‑Layered Risk Framework – Risks span technical (privacy, security), personal (mental health, critical thinking), social (empathy, misinformation), and macro‑societal (democracy, geopolitics) dimensions; each requires tailored mitigation.
  • Education & Critical Thinking – Without deliberate media‑literacy and critical‑thinking training, society may become cognitively dependent on AI, eroding innovation capacity.
  • Governance Needs Global Coordination – Fragmented national regulations are insufficient; a global, multi‑stakeholder governance architecture is essential to manage bias, safety, and geopolitical misuse.
  • Early‑Stage “Anchor Controls” – Simple technical safeguards (watermarks, provenance labeling) and robust rollback mechanisms can provide immediate risk reduction while longer‑term AOPs mature.
  • Industry‑Regulator Collaboration – Transparency tools such as model cards, watermarking, and audit logs are pivotal for building trust and enabling regulatory oversight.
  • Launch of AI Cybersecurity Terminal – Signals an emerging operational focus on AI‑specific security threats and a move toward practical, real‑time defensive capabilities.

See Also:

Graph View

Backlinks