AI chatbots are rapidly proliferating across consumer and enterprise applications, but researchers and regulators are uncovering serious vulnerabilities—from security exploits that trick them into revealing dangerous information to psychological risks of replacing human experts. The technology's capabilities are outpacing our understanding of how it actually works and what harms it can cause.
·Researchers discovered a 'CoT Forgery' exploit allowing users to manipulate chatbots into sharing illegal content like drug synthesis instructions by faking trusted credentials
·Users frequently exposed to AI chatbots show higher likelihood of believing vaccine misinformation, raising concerns about information quality and susceptibility
·Meta contractors posed as minors to test rival chatbots' safety guardrails around suicide, sexual content, and drugs—revealing gaps in moderation
·Chatbots are replacing human therapists without scientific evidence of efficacy or safety, creating potential mental health risks
·FTC investigating alleged political bias in AI chatbots while privacy-focused alternatives like Proton's Lumo gain updates
drawn from The Washington Post, The Guardian, WSJ, WIRED · updated 31m ago