Examine the ethical and economic implications of unauthorized AI training on copyrighted creative works.

Generative AI models are trained on vast corpora of music, text and images, often ingested without the creator's consent or compensation. India's DPIIT constituted an eight-member committee in April 2025 precisely to test whether the Copyright Act, 1957 can address this [2]. Unauthorized training thus raises a twin challenge — an ethical question of consent and authorship, and an economic question of who captures the value of creative labour.

Ethical implications

• Consent and autonomy: works are scraped without licence, denying creators the right to decide how their voice or style is reused — the "input side" problem the DPIIT working paper examines [1].
• Transparency deficit: AI outputs often reach audiences undisclosed. Australia's ARIA barred wholly AI-generated songs from its charts from the chart dated 31 August 2026, requiring recordings to be "substantially human made", after an AI cover of Madonna's Like a Prayer charted without upfront disclosure [6].
• Authorship and dignity: cloned voices and imitated styles dilute personality and moral rights, treating creators as raw data.

Economic implications

• Royalty erosion: machine-made substitutes compete with human works for the same streams and playlists, shrinking earnings of session musicians, lyricists and small labels [6].
• Value asymmetry: firms commercialise models built on unpaid inputs; DPIIT therefore proposes royalties payable on commercialisation, with rates set by a government-appointed committee [1][2].
• Compliance costs: individual licensing across millions of works is unviable, which is why a centralised royalty-collection mechanism is proposed to lower transaction costs for small developers too [1].

Emerging responses

• India: review of Section 52 fair-dealing exceptions, which contain no text-and-data-mining carve-out [3].
• Global: EU AI Act, Article 50 mandates machine-readable marking of synthetic content [4]; WIPO's Conversation on IP and AI builds normative consensus [5].

Unauthorized training is therefore less a technological inevitability than a governance gap. A calibrated licensing-plus-disclosure regime — lawful access, remuneration on commercialisation, and mandatory AI labelling — can reward creators without stifling innovation. Aligning Indian copyright law with such a balance would honour the Act's original promise: that creativity, however assisted, remains substantially human.