India considers new law to rein in AI risks
What's the story
India is considering a new artificial intelligence (AI) law that could create a separate safety framework for some of the world's most powerful AI models, as per NDTV. The proposed legislation would require developers to prove that their systems have adequate safeguards in place before they can be released into the public domain. This move is aimed at addressing risks posed by increasingly capable AI systems, including those that can find software vulnerabilities or enable sophisticated cyberattacks.
Upcoming release
Consultation paper on AI law to be out soon
Union Electronics and IT Minister Ashwini Vaishnaw has announced that a consultation paper on the new AI law will be released in the coming weeks.
The document is expected to seek input from civil society, citizens, industry players, and other stakeholders.
A major point of contention is whether India should differentiate between regular AI applications and "frontier models," which are at the cutting edge of capability but could pose greater risks if misused.
Global influence
US AI regulation could serve as a model
The US experience with AI regulation could serve as a model for India.
In June, restrictions were placed on Anthropic's Mythos 5 and Fable 5 models over national security concerns. However, these restrictions were lifted later in July.
The Indian government is likely to discuss pre-release assessment, controlled environment testing, and safeguards proportional to a model's capabilities while drafting the consultation paper.
Testing requirements
AI firms may need to test models in a sandbox
One possible requirement being considered is that developers of particularly capable models may have to test them in a sandbox before wider deployment.
This would allow researchers to assess a model's behavior and identify vulnerabilities before its release.
The government is also likely to explore whether developers should limit access to certain functions, restrict certain users, or deploy different safeguards depending on the risks associated with a model.
Regulatory challenges
What will be the 'frontier model' definition?
A key challenge will be defining what constitutes a "frontier model." A threshold based on computing power used in training could be one way, but capability-based assessments may also be required.
The proposed law could extend beyond regulating content distributed through digital platforms to examine the systems that generate it.
This distinction is important because content moderation alone may not address risks such as autonomous AI agents taking unauthorized actions or models being used to identify vulnerabilities in critical infrastructure.
Risk mitigation
Proposed framework likely to address deepfakes, financial fraud risks
The proposed framework is unlikely to be limited to frontier models. It will likely address issues like deepfakes, financial fraud, impersonation, psychological harm, and the growing difficulty of distinguishing authentic material from AI-generated content.
As per NDTV, officials have indicated that "deepfakes," "financial frauds," and "psychological harms" are among the areas that could be addressed in the consultation for this new law.