Government Leaves AI Chatbot Safe Harbor Unsettled
The government says eligibility depends on what an AI service does but offers no separate test for systems that generate content.
Topics
News
- Apple Tightens Mac Access Rules as AI Agents Gain Reach
- Trump Puts Intelligence Chief Jay Clayton in Charge of AI Task Force
- IBM Widens AI, Quantum Research With IIT Bombay, IISc
- OpenAI, Synopsys Build AI Model That Could Slash Chip Design Time
- Pentagon Plans Four-Star Command for AI, Drone Warfare
- California Draws the Line on AI Firings, Layoffs and Workplace Surveillance
India has stopped short of deciding whether chatbots and other generative AI services qualify for intermediary safe harbor, saying the answer depends on what each service does.
Congress member of Parliament Manish Tewari raised the issue in the Lok Sabha on Wednesday, July 29. He asked whether generative AI systems, including chatbot services, fall within the definition of an intermediary under the Information Technology Act and can claim protection under Section 79.
He also asked whether the government would distinguish between services that provide access to third-party information and those that generate new content.
Minister of State for Electronics and Information Technology Jitin Prasada said the IT Act was technology-neutral and applied whether or not a service used AI. Whether a system qualified as an intermediary under section 2(1)(w) would depend on the service provided, the functions performed and the applicable provisions of the act and its rules, he said in a written reply.
The answer set out no test for telling a content-generating AI service apart from a platform that merely hosts or transmits information. Nor did it announce separate rules for generative AI providers.
Safe harbor is the legal shield that stops a platform from being held liable for what its users post, provided it meets due-diligence duties. Section 79 of the IT Act grants that protection to qualifying intermediaries for third-party information. The protection is conditional. Among other requirements, an intermediary must not initiate a transmission, select its recipient or select or modify the information it carries.
That framework was written for services handling information supplied by others. Generative AI unsettles the distinction because it builds new responses from user prompts, training data and the model’s own processing.
The government said it was consulting stakeholders on the legal, policy and institutional questions raised by emerging technologies. Any further legislation or regulation, if considered necessary, would follow the established process.
It also pointed to the Information Technology Rules amended in February. Those rules require intermediaries that enable the creation or distribution of synthetic media to use reasonable technical measures against unlawful material. Other synthetic media must carry prominent labels and, where technically feasible, permanent provenance metadata.
The rules, though, define synthetically generated information as audio, visual or audiovisual material made to appear real. MeitY’s accompanying clarification says the definition does not cover written text. The amendments therefore reach deepfakes, cloned voices and realistic synthetic images or video, but leave the status of chatbot text unsettled.
The amendments also cut the time to comply with a valid court order or government direction to remove unlawful content, from 36 hours to three. A tighter two-hour window applies to non-consensual intimate imagery and deepfakes. The deadlines apply to content generally, not only to AI-generated material.
The legal uncertainty has already reached the courts. In May, the Calcutta High Court called the classification of ChatGPT as an intermediary or an originator a “complicated and vexed” question that would need technical and expert evidence at trial.
At the preliminary stage, the court said ChatGPT’s generative functions made it look closer to an originator than a passive intermediary. It added that the issue would ultimately require legislation dividing responsibility among AI developers and users according to their control over the system.
The court made those observations while dismissing an interim application by IndiaMART against OpenAI. The underlying case remains pending.
