In February, the generative AI startup Anthropic refused, for ethical reasons, to grant unlimited access to its AI model Claude to the Pentagon. The firm demanded that a series of red lines be respected regarding the use of its technology in mass surveillance of American citizens and autonomous weapons. Besides labelling it as “radical left” and “woke”, Donald Trump ordered the cancellation of all its contracts with the Department of War and included it on a list of companies deemed dangerous to national security.
Anthropic is a rare bird in the universe of AI. Its founding manifesto advocates for a safe, responsible, and positive AI for humanity. Its CEO, Dario Amodei, left OpenAI in 2021 precisely because he believed it did not meet safety standards, and established his new company as a public benefit corporation: legally obligated to build models that respect human rights. Amodei has opened up to conversations with the White House but has accused it of violating the First Amendment with its veto on Anthropic.
At the other end of the discourse is Palantir, another of the major companies in the sector. Its manifesto, a 22-point text published in April on the social network X, has been unanimously described as an example of “technofascism”. In contrast to Amodei's warning that a misuse of such a powerful technology as AI could lead to the derailment of democracy, Palantir's CEO, Alex Karp, believes it is the main weapon to ensure the survival of the West. Bureaucracy and democracy hinder progress, and Silicon Valley has an almost moral duty to participate in the defence of the nation. The use of its technology by ICE to identify and locate migrants in the US and its collaboration with the Israeli Ministry of Defence is well documented.
Amnesty International places the US, China, Pakistan, Afghanistan, South Sudan, and Tanzania in the catalogue of countries that by 2025 have applied authoritarian practices through technology such as the unlawful surveillance of citizens to limit the right to freedom of expression or suppress protests. The regulation of AI remains insufficient. For example, the AI Governance Guidelines developed in India were non-binding or very general.