Global Witness

Global Witness is a London-based international accountability NGO founded in 1993 by Patrick Alley, Charmian Gooch, and Simon Taylor. Its primary mandate is exposing the connections between natural resource exploitation, environmental destruction, and human rights abuses — tracking corporate malfeasance in conflict minerals, deforestation, fossil fuels, and illicit financial flows. Through its Digital Threats programme, Global Witness has produced a series of original platform accountability investigations exposing how social media companies' AI ad review and content moderation systems systematically approve disinformation in violation of the platforms' own stated policies. This adversarial investigation work places Global Witness in the movement as an investigative infrastructure organization documenting AI harms.

Digital Threats programme and adversarial testing methodology

Global Witness's Digital Threats work centres on adversarial testing of platform AI content moderation and ad review systems. The methodology involves creating content that explicitly violates platform policies — disinformation, incitement, election interference — submitting it through normal advertiser channels, and documenting whether AI systems catch it. The investigations produce both failure documentation (approval receipts, screenshots, timestamps) and analysis of why specific content categories escape algorithmic detection. By publishing results alongside the violating content itself, Global Witness creates a record of AI moderation failure that platforms cannot dismiss as theoretical.

AI-generated election disinformation investigation (2023)

In September 2023, Global Witness published an investigation documenting that TikTok and YouTube approved every one of 100 AI-generated election disinformation ads submitted to their platforms. The ads contained false claims about voting procedures, election officials, and polling locations — content explicitly targeting the integrity of electoral participation. Both platforms' AI ad review systems approved all submitted content despite policies nominally prohibiting election integrity violations. The investigation was published ahead of what Global Witness identified as a 2024 election cycle spanning more than 50 countries, with a combined electorate facing AI-generated disinformation at scale. Coverage by the Guardian, BBC, and major outlets made the failure documentation widely visible in pre-election policy debates.

Tigray conflict disinformation investigation (2021)

In 2021, Global Witness tested Facebook's AI ad review system against content related to the conflict in Ethiopia's Tigray region. Ads containing false claims and incitement content that violated Facebook's stated policies were submitted through the platform's ad creation interface. Facebook's AI review system approved the content. The investigation documented the specific policy violations each ad contained alongside approval records — establishing that the failure was not an edge-case ambiguity but a systematic breakdown of AI moderation against content directly relevant to an active humanitarian crisis. Global Witness published the findings with enough specificity (content descriptions, approval evidence) that the investigative record could be independently evaluated.