Science and Technology › Computing and AI

AI governance, safety and ethics

How artificial intelligence is governed: national and international rules, safety and the risks of frontier models, deepfakes, copyright, bias, liability and autonomous weapons. Prelims has asked about the AI Action Summit in Paris.

UPSC has asked

Prelims 2025: the AI Action Summit held in Paris

Showing 2 of 8 articles, those that changed from 1 to 30 June 2026.Show all

Export controls on frontier models

Mains

Export controls on artificial intelligence are curbs on what a country lets out, and they work on scarce hardware such as advanced chips.

A model whose weights have been published cannot be withdrawn, which limits what controls on models themselves can do.

US order versus India's 2024 advisory

	US executive order, 2026	India's advisory, March 2024
Nature	A voluntary testing window of up to 30 days before release	Permission required before launch
Scope	Covered frontier models	All untested AI
Fate	In force when reported in June 2026	Withdrawn a fortnight later

The US chose a voluntary window for the most capable models, where India's advisory had asked permission for all untested AI and did not survive industry pushback.Source: The Indian Express, 3 June 2026

What changed

16 Jun 2026
- Talks between Anthropic and US officials reached no resolution; the government's concern is reported to be a narrow jailbreak that let a user bypass a cybersecurity guardrail.
- India had secured narrow access to Mythos class models through Project Glasswing.
- India spends under 0.7 per cent of GDP on research and development, against over 2.5 per cent in China and over 3 per cent in the US.
Mains: India should not seek self sufficiency across the whole AI value chain but identify which dependencies create strategic exposure, concentrating on compute for sensitive workloads, cyber defence and targeted frontier capabilities; access to frontier AI should be decided under a transparent, globally negotiated framework with third party testing before release.
The Indian Express, 16 Jun 2026: Anthropic and US govt officials hold talks as AI export controls remain in place (opens in a new tab) · The Indian Express, 19 Jun 2026: Anthropic blackout: This is what strategic AI dependence looks like (opens in a new tab) · The Indian Express, 20 Jun 2026: Anthropic blackout episode shows that frontier AI needs global governance (opens in a new tab) · The Hindu, 18 Jun 2026: Innovate or be eaten: On India and an innovative ecosystem (opens in a new tab)

Show history (1 other update)

3 Jul 2026BriefA Chinese firm released a rival open weight model within a day of the restriction; controls work on scarce hardware, but published model weights cannot be withdrawn. The Hindu, 3 Jul 2026: Can caged frontier AI models outrun open ones? (opens in a new tab)

AI in courts: the draft 2026 Regulations

Prelims and Mains

The draft Regulations for Use of Artificial Intelligence in Courts, 2026, prepared by the Supreme Court's AI Committee, are a 57 regulation framework for every court and tribunal in India.

Regulation 4 makes AI strictly assistive and subservient to human judgement.
Regulation 20 lists prohibited uses as absolute and non derogable, including risk scoring for flight risk, recidivism, bail eligibility or witness credibility; Regulation 56 bars the Apex Body from relaxing it.
Permitted uses include transcription, translation, legal research, scheduling, accessibility services and chatbots.
The ban traces to the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS), a US recidivism tool found to misclassify Black defendants, and SyRI, a Dutch welfare risk scoring system struck down in 2020 on privacy grounds.

Oversight chain

Supreme Court Apex Body
Sets the framework and may not relax the prohibited uses
High Court AI Secretariats
Keep the AI Register and Incident Database, run audits and publish transparency reports
High Court AI Committees
Hear grievances from litigants about AI use in their cases

Oversight runs down from the Supreme Court to every High Court, and a litigant's complaint runs up through the same chain.Source: Frontline, 16 June 2026

What changed

16 Jun 2026New
- The draft was released on 3 June; about 52 million cases were pending across the judiciary as of mid 2026.
Mains: The bans are strong on paper but the oversight machinery is unfunded and beyond the capacity of district courts, and a High Court AI Secretariat must judge vendors' own descriptions of their tools, so a risk tool could be rebranded as a scheduling aid.
Frontline, 16 Jun 2026: A firewall against algorithmic justice (opens in a new tab)

Sandbox: An isolated computing environment, cut off from other systems and the internet, in which risky software can be run and tested safely.

Risk based regulation: Rules scaled to the harm a technology could cause, so the strictest duties fall on the riskiest uses.

Conformity assessment: The process of showing that a product meets legal standards before it is put on the market.

Watermark: A hidden, machine readable signal embedded in content to show that a machine generated it.

Synthetically generated information: Audio, image or video artificially or algorithmically created or altered so that it appears real and is likely to be taken for a real person or a real event.

Metadata: Data carried alongside a file that records where it came from and how it was made.

Data sovereignty: A country seeking control over data originating within its borders.