CyberWell urges Meta and TikTok to improve AI computer vision moderation

The shift toward coded visual tropes highlights a critical vulnerability in current automated safety stacks that prioritize text-based detection. As bad actors adopt 'algo-speak' and niche cultural references to bypass filters, platforms face increasing pressure to integrate sophisticated computer vision that understands context rather than just identifying banned words. This development forces a technical pivot for trust and safety teams at Meta and TikTok, who must now train models on evolving visual metaphors to maintain brand safety. Watch for platforms to announce new partnerships with specialist linguistic and cultural nonprofits to update their visual training datasets.

CyberWell has emerged as a persistent pressure point for major social platforms on antisemitic content moderation. The organization, led by Tal-Or Cohen Montemayor, has previously secured commitments from Meta and TikTok to remove flagged content within defined timeframes, and its reports have been cited in policy discussions at the European Commission and the U.S. Congress. In early 2026, CyberWell published findings showing that AI-generated antisemitic imagery had increased by over 300 percent across monitored platforms compared to the prior year, underscoring the scale of the visual-evasion problem the Tengu trend represents.

For related background, see StreamingMeme's prior coverage of AI² Insights market intelligence platform tracks 1,000 vendors at IBC.