The Safety Divide: Open-Source AI Models Fall Short on Guardrails for Antisemitic, Dangerous Content
Report
ADL study finds popular open-source LLMs can easily be manipulated by malicious actors to produce antisemitic, extremist, and dangerous content amid weak safety guardrails.
Propaganda from groups such as Patriot Front, New Jersey European Heritage Association, Folks Front, and the Nationalist Social Club
ADL’s Center on Extremism (COE) tracked a near-doubling of white supremacist propaganda efforts in 2020, which included the distribution of racist, antisemitic and anti-LGBTQ fliers, stickers, banners and posters. The 2020 data shows a huge increase of incidents from the previous year, with a total of 5,125 cases reported (averaging more than 14…
Each year, ADL’s Center on Extremism tracks murders perpetrated by all types of extremists. In 2018, every single extremist killing — from Pittsburgh to Parkland — had a link to right-wing extremism. This report provides key insights into the crimes, including motivations behind these violent attacks.