Information Technology Act, 2000
AI-Generated vs Human-Generated Content: Law, Liability and Proof
Content law grew up assuming a human on the other end of every sentence and image; generative models broke the assumption without breaking most of the law, because the offences ask what content does, not who made it. The comparison therefore runs on four working axes, ownership, liability, labelling and proof, and on one honest admission: the authorship question for machine output is not yet settled. Topics 92 and 112 built the AI framework; this note, as asked, is the dedicated comparison.
1. The Comparison
Figure 1: Model output against authored work
- Ownership. Human work vests copyright on creation, the author owning, licensing and suing on it. AI output sits in the gap: the Copyright Act's computer-generated works clause names as author the person who causes the work to be created, yet originality doctrine presupposes human creativity, so purely machine-generated output is on the prevailing view unprotected, with meaningful human prompting, selection and arrangement argued as the route to authorship, and the question unsettled (Topic 92)
- Liability. For human content the author and publisher answer directly. For AI output liability climbs the accountability chain: the user who deploys or publishes it answers as if they had written it, the enterprise deploying a chatbot answers for its promises and misstatements, and the tool's provider answers for its own statutory duties and, on ordinary principles, where it participates in the wrong; the model itself is no legal person and answers for nothing.
- Labelling. Human content owes no disclosure of its making. AI content that appears authentic is synthetically generated information, so the 2026 duties attach, tool labels, upload declarations, SSMI verification and display (Topics 75, 112)
- Proof. Both are electronic records proved under the BSA with the s.63 certificate; the AI-generated exhibit adds provenance questions, which model, whose prompt, what pipeline, that certification and expert examination must carry, and the s.79A examiner increasingly answers (Topics 9, 82)
2. The Neutral Offences and the Open Questions
Figure 2: Who answers along the chain
- Technology-neutral offences. An AI-written phishing message is s.66D and cheating against its sender exactly as a typed one; AI-generated intimate imagery is s.66E and the content offences; AI-assisted forgery is forgery. The generation step neither creates an offence nor excuses one, the principle that disposes of most AI liability questions (Topic 92)
- Defamation by model output. A hallucinated allegation published is defamation by its publisher: the person or business that put the output before readers, with the tool provider's exposure turning on its role and knowledge; disclaimers reduce no one's liability for content they chose to publish.
- Misinformation at scale. Model-written falsehoods meet the same design as manipulation generally, the due diligence categories, neutral offences, courts and counter-speech, with the SGI label disclosing origin but no organ adjudging truth (Topics 81, 112)
- What remains open. Copyright in AI output and liability for training on protected works are unsettled or sub judice; the intermediary status of generative tools for what they produce is untested; and no Indian statute yet addresses algorithmic authorship as such, the honest frontier every good answer marks (Topic 92)
⚠ Exam trap Organise by the four axes, ownership, liability, labelling, proof, and hold the two anchors: the offences are technology-neutral, so the human who deploys AI output answers for it as for their own words, and the authorship question is unsettled, purely machine output likely unprotected while human selection and arrangement is the argued route in. Attach the labelling duty to appearance of authenticity under the SGI regime, not to AI use as such, and in evidence answers add the provenance layer, model, prompt and pipeline, that an AI exhibit's certificate and examination must address. |
3. Frequently Asked Questions
Who is liable for harmful AI-generated content?
The human or entity that deploys or publishes it. The offences are technology-neutral: an AI-written scam is Section 66D and cheating against its sender, AI-generated intimate or obscene imagery engages Sections 66E and 67 onward against those who make and circulate it, and a hallucinated defamatory statement is defamation by its publisher. The tool's provider answers for its own duties, SGI labelling and due diligence among them, and on ordinary principles where it participates in the wrong; disclaimers do not transfer liability to the machine, which is not a legal person.
Does copyright protect AI-generated content in India?
Uncertainly, and often not. The Copyright Act's computer-generated works clause designates the person who causes the creation as author, but originality doctrine presupposes human creative contribution, so purely machine-generated output is generally regarded as unprotected; substantial human prompting, selection and arrangement is the argued basis for authorship in mixed creations. Human-generated work, by contrast, vests copyright on creation, and the training-data question, whether models may learn from protected works, is sub judice.
4. Related Topics
- Topic 92: AI and Indian Cyber Law. The whole AI framework.
- Topic 112: Synthetic vs Manipulated Content. The labelling line in detail.