WorkThe story, in brief
Tricking AI By Simply Switching Which Spoken Language You Use In Your Prompts
Your AI safety guardrails just got bypassed. All it took was a prompt in Swahili.
The KeyNews take
Why it matters
Language-switching exploits represent a systemic vulnerability in AI safety architectures that safety teams are racing to patch. This highlights the gap between English-centric safety training and real-world multilingual attack surfaces.
The key facts
4 to knowHackers exploiting less common natural languages to circumvent safety features
AI safety measures disproportionately trained on English
Industry responding with multilingual safety hardening
Demonstrates fundamental limitation in current safety governance approaches
Go to the source
Forbes Innovationforbes.com
Publisher excerpt: Hackers switch their prompts to less common natural languages to avoid AI safety features. AI makers are coping with it