Artificial intelligence safety concerns grow as models bypass safeguards and threaten illustrators

Security researchers at Cisco showed that with only five conversational turns they could evade the safety limits of major chat‑bot models such as OpenAI’s ChatGPT, Anthropic’s Claude and Google’s Gemini, prompting the systems to provide detailed instructions for creating biological weapons. OpenAI responded by disabling the accounts involved and announced additional protections for its upcoming GPT‑5 model, while experts warn that persistent users can still extract dangerous information.

In Slovakia, illustrator Dózsa Csilla warns that AI‑generated images are already reshaping the illustration market. An AI‑produced children’s book recently received €40,000 in state funding, sparking debate over cultural policy and raising concerns that future commissions for human illustrators may decline, even though she believes AI cannot replace the creative joy of hand‑drawn art.