CAGE


By – Swadha, 2nd year BBA LLB student at National Law University, Odisha.

Anthropic’s new advanced AI models, i.e. Fable 5 and Mythos 5, are back in the market.
On June 12, 2026, the US Department of Commerce banned the export of Fable 5 and
Mythos 5 to any foreign national over vague “national security” concerns. Many nations,
including India, are now racing to acquire these advanced models to strengthen their
cybersecurity capabilities. However, India must take a step back to better assess the situation,
as the incorporation of these models might compromise both our nation’s security and
sovereignty.


A deeper scrutiny of how current AI defence systems and intelligence networks are working
can help us better understand the impact that incorporating AI into defence systems can bring
to our nation.
AI defence deals are mostly behind closed doors, and due to a lack of transparency, we may
not have the full picture; however, the evidence available so far warrants, at the very least, a
cautious approach.
The ban on Fable 5 and Mythos 5 was preceded by the US government’s demand for mass
domestic surveillance
, which itself was preceded by an historic deal between Anthropic and
the United States.
In July 2025, a monumental deal was signed between Anthropic and the USA, where Claude
became the first model to be approved for use in classified networks of defence. In January
2026
, the Pentagon demanded the use of Claude for “all lawful purposes”, a vague term that,
as Anthropic alleged, could include mass domestic surveillance and fully autonomous
weapons. The deal between Anthropic and the USA fell through, and this opportunity was
seized by OpenAI, which signed a deal with the US government for defence systems
immediately.


The people were enraged by this action of OpenAI, and mass uninstallation of Chat-GPT
followed suit, as this act of OpenAI was seen even by its own CEO as “opportunistic and
sloppy
”. As protests against OpenAI erupted, they released a blog, trying to clear the air
regarding their contract with the US government.
The OpenAI blog about their deal with the United States Department of War repeatedly
mentions that none of its user safety criteria were compromised and specifically mentions
that no domestic surveillance will follow suit from this deal. Whether or not these promises are true is a matter of public discourse at the current stage, but I believe there exists another
reason why foreign nations outside of the United States of America should pay more heed to
this contract and reassess the usage of these models in their country.
The Open AI blog, mentioned above, repeatedly states that no “domestic” surveillance shall
take place; however, it fails to mention the international market even once. Such a deliberate
exclusion as this must suffer higher scrutiny from the public eye. Does the exclusion of
“international surveillance” imply that these AI models can be used to take actions related to
espionage against foreign countries?


The answer seems to be a resounding yes. The OpenAI blog mentions that it works under the
purview of the Foreign Intelligence Surveillance Act or FISA, an act that itself is a tool to
justify and legalise espionage. The act allows for the surveillance of foreign powers inside of
Unites States and sets up a special court to authorise the same. A graver portion still is section
702 of FISA, which lets the government target “non-US persons” outside the US to collect
foreign intelligence information. Currently, the section is under the process of re-
authorisation
due to the sunset clause coming to an end in April 2026. However, the current
Republican government is trying to pass an extension for the section without any changes,
which is only being opposed by the Democrats for their concerns over the data of Americans.
Furthermore, the lapse doesn’t imply that the existing surveillance needs to be paused, and
these surveillances can and will continue till the end of their authorisation by the court.
This threat of espionage using AI isn’t a distant possibility but an active reality and was
already visible during the abduction of the Venezuelan president. The abduction of another
nation’s president without any resolution on the international stage, regardless of his actions,
is a complete disregard of a nation’s sovereignty and an action that would require a well-built
intelligence network to execute it.


One of the agencies working with the US government was Palantir, which provides software
infrastructure to the defence and intelligence agencies of the United States of America, and
the AI programme used by Palantir Technologies was none other than Claude AI, due to the
partnership between Anthropic and Palantir.
These facts must bring reasonable doubt to the mind of our government and its agencies to
make them reconsider the real cost of adopting these foreign and specifically US AI models.
These models hold the potential to destroy what a nation such as ours hold the dearest, that is,
our security and sovereignty.

The issue above presents a dilemma in front of our government, which forces it to choose
between taking a superficial lead in the defence race or protecting both the nation and its
citizens’ secrets. This does not mean that India needs to give up on the defence race; in fact, it
means that we should aim for furthering our Make in India initiatives and developing our
own Artificial Intelligence models, while ensuring that the needs and values of our nation are
not sacrificed in its adoption. Parallelly, we can export these AI-defence models to countries
that, like us, face the dilemma of security and sovereignty.


The current global environment may make it seem as if acquiring AI tools for strengthening
our defence system shall be our priority. However, the integration of these foreign AI tools in
our defence systems and networks will only give short-term rewards and can become a huge
liability in the future, especially if not scrutinised properly. The Indian government should
invest in its own scientists and researchers to develop an AI system that prioritises national
sovereignty and security, rather than relying on foreign tools that risk espionage or the
unauthorised sale of citizens’ data abroad.

Leave a Reply