From: https://brownstone.org/articles/did-ai-almost-start-world-war-iii/

As DD Geo-politics reported, “since 2015, the IAEA has relied on Palantir’s Mosaic platform, a $50-million AI system that sifts through millions of data points — satellite imagery, social media, personnel logs — to predict nuclear threats.”

In this particular case, reports Alastair Crooke,

“Its algorithm looks to identify and infer ‘hostile intent’ from indirect indicators — metadata, behavioral patterns, signal traffic — not from confirmed evidence. In other words, it postulates what suspects may be thinking, or planning. On 12 June, Iran leaked documents, which it claimed showed IAEA chief Rafael Grossi sharing Mosaic outputs with Israel. By 2018, Mosaic had processed more than 400 million discrete data objects and had helped impute suspicion to over 60 Iranian sites such as to justify unannounced IAEA inspections of those sites, under the JCPOA. These outputs, though dependent largely on the algorithmic equations, were incorporated into formal IAEA safeguard reports and were widely accepted by UN member states and non-proliferation regimes as credible, evidence-based assessments. Mosaic however is not a passive system. It is trained to infer from its algorithm hostile intent, but when repurposed for nuclear oversight, its equations risk translating simple correlation into malicious intent.”