US v China: What if an AI system started a war that no one wanted?
At the start of the US war against Iran, the US Air Force used intelligence synthesised by Project Maven to select targets in deadly bombing raids. Project Maven is led by Palantir and supported by American technology behemoths like Amazon and Microsoft. Its use led to the bombing of the Minab school in Iran’s southern Hormozgan province, which killed more than 150 people in an incident described by experts as a war crime.
Meanwhile, militaries around the world are also quickly trying to acquire and adapt Ukrainian drone technology, which this year marked the first use of a lethal autonomous system in battle.
Outside of military conflict, AI models pose a threat to national information systems and critical infrastructure. The recent breach of Australia’s Medicare system exposed a new threat vector in critical systems: that a hack is no longer just the domain of foreign intelligence services or cyber criminals. A hack can now happen accidentally, without human control, emerging from a friendly country. This raises the possibility that such an accident might happen between the US and China, which together control 90% of global AI capacity.
In the first documented instance in which OpenAI hacked Australian government servers, OpenAI itself was unaware of the breach for two months and then waited another month to inform the Australian government via a public e-mail.
posted by Dark Shadow | 1:01 PM
