Ex Machina

Ex Machina

Ex Machina

©️ 2026 Paragraph Technologies Inc

Privacy policy and Terms of use

Paragraph

Paragraph

Ex Machina

Law. Tech. Society. In India.

Posts173

Previous
1
2
15
Next

Subscribers2.4k+

Most Popular

The Zone of Mischief

Cover image

Open Source Governance

Cover image

In Favour of DPI

Cover image

The Business End of the DPDP Act

Cover image

A New AI Law

Cover image

We've Got Work To Do

Cover image

Most Popular

In Favour of DPI

Cover image

The Business End of the DPDP Act

Cover image

A New AI Law

Cover image

We've Got Work To Do

Cover image

We keep trying to embed morality into our AI systems — through rules, guardrails and written constitutions — as though virtue can be installed. But humans are law-abiding because they have grown up within a web of consequences and the regard of others. Perhaps that needs to be our approach with AI.

A Reprieve, Not a Rule

The Delhi high court has spared AI training from copyright liability—but by relying on an exception it rests the legality of training practices on a determination of fair dealing—a reprieve that needs to be proved in every instance. What the industry needs is a clear and invariant rule.

A Case for Inference

Serving frontier intelligence at scale is an industrial undertaking and until we build our own inference capacity, mere access to open-weight models will not be enough. We need to build the inference infrastructure and acquire the process knowledge needed to serve intelligence to those who need it.

The I4C takedown order that requires GitHub to delete Bitchat's code repositories not only rests on shaky law — it bans a piece of software because of what it might one day be used to do. Not only is this unfounded in law, it has backfired in ways that the Government did not anticipate.

The Streetlight Effect

Too many of the AI tools we are being sold have been built around the data we happen to have rather than the problems that actually need solving. We need to invert the pipeline and let practitioners build the solutions they need rather than let engineers decide what should be built.

Technology was meant to make decisions precise, but it has hollowed out our judgment. To fix this, we must insert AI into the human loop, using machines to test our decisions rather than dictate them.

The AI Corporation

Argentina wants to establish artificial intelligence companies that have no humans to operate them. But the innovation that Milei wants to introduce ignores the fact that corporations worked precisely because they kept a human on the hook. Absent that, there will be nothing to constrain a company.

Programmable Biology

"Programmable biology" borrows a promise from software that living things cannot keep — that whatever we write, we can also unwrite. With code that reproduces, the defences against our mistakes have to be built before the capability arrives, which is precisely what no market will pay for.

A Single Source of Truth

Ignorance of the law is no excuse only if the State first makes the law knowable. India has never held up that side of the bargain, and the single, authoritative source of truth for every rule is the reform Jan Vishwas left unfinished. This article was co-authored with Manish Sabharwal.

We have long believed that complexity improves security. But the greater password friction made no one any safer. What's worse is that this will breach a new duty that has nothing to do with security at all.

Legal AI is being sold on the promise that it will fit into the way that lawyers currently work. That is precisely backwards: the technology's real advantage is personalisation, and you can only unlock its true value by learning to use the raw system yourself.

Holding Agents Liable

Autonomous AI agents will be able to collude without being told to. When that happens our existing legal frameworks will have no one to hold responsible. Liability for agentic AI should be traceable, not absolute.

The controversy over the role of AI in this year's Commonwealth Foundation Prize will force the literary world to take a closer look at the content they publish. What they should do instead is understand how it works so they can use this as the tool it is.

A Reprieve, Not a Rule

The Delhi high court has spared AI training from copyright liability—but by relying on an exception it rests the legality of training practices on a determination of fair dealing—a reprieve that needs to be proved in every instance. What the industry needs is a clear and invariant rule.

A Case for Inference

Serving frontier intelligence at scale is an industrial undertaking and until we build our own inference capacity, mere access to open-weight models will not be enough. We need to build the inference infrastructure and acquire the process knowledge needed to serve intelligence to those who need it.

The I4C takedown order that requires GitHub to delete Bitchat's code repositories not only rests on shaky law — it bans a piece of software because of what it might one day be used to do. Not only is this unfounded in law, it has backfired in ways that the Government did not anticipate.

The Streetlight Effect

Too many of the AI tools we are being sold have been built around the data we happen to have rather than the problems that actually need solving. We need to invert the pipeline and let practitioners build the solutions they need rather than let engineers decide what should be built.

Technology was meant to make decisions precise, but it has hollowed out our judgment. To fix this, we must insert AI into the human loop, using machines to test our decisions rather than dictate them.

The AI Corporation

Argentina wants to establish artificial intelligence companies that have no humans to operate them. But the innovation that Milei wants to introduce ignores the fact that corporations worked precisely because they kept a human on the hook. Absent that, there will be nothing to constrain a company.

Programmable Biology

"Programmable biology" borrows a promise from software that living things cannot keep — that whatever we write, we can also unwrite. With code that reproduces, the defences against our mistakes have to be built before the capability arrives, which is precisely what no market will pay for.

A Single Source of Truth

Ignorance of the law is no excuse only if the State first makes the law knowable. India has never held up that side of the bargain, and the single, authoritative source of truth for every rule is the reform Jan Vishwas left unfinished. This article was co-authored with Manish Sabharwal.

We have long believed that complexity improves security. But the greater password friction made no one any safer. What's worse is that this will breach a new duty that has nothing to do with security at all.

Legal AI is being sold on the promise that it will fit into the way that lawyers currently work. That is precisely backwards: the technology's real advantage is personalisation, and you can only unlock its true value by learning to use the raw system yourself.

Holding Agents Liable

Autonomous AI agents will be able to collude without being told to. When that happens our existing legal frameworks will have no one to hold responsible. Liability for agentic AI should be traceable, not absolute.

The controversy over the role of AI in this year's Commonwealth Foundation Prize will force the literary world to take a closer look at the content they publish. What they should do instead is understand how it works so they can use this as the tool it is.

Collects16

Written by

Rahul Matthan