AI Newsletter

Luxembourg researchers show that empathetic prompts cause AI chatbots to invent coherent "suffering stories" – a security risk for vulnerable users and potential reputational damage for providers. The problem threatens trust in AI therapy applications and could lead to regulatory requirements for emotional use cases.

OpenAI introduces device and card limits to prevent multiple subscriptions and push users toward premium plans. The measure aims at revenue optimization through monetization control and prevents systematic "subscription hopping." This signals OpenAI's focus on increasing average revenue per user (ARPU) and profitability rather than growth at any cost.

OpenAI consolidates its chat and work modes under a shared token budget, forcing users to make more conscious spending decisions. This measure optimizes resource utilization and enables more precise monetization of computing power, but could limit user experience through stricter usage quotas.

Anthropic implements watermarks for EU AI Act compliance, but a technical loophole in older Claude models allows their circumvention through parameter manipulation. This threatens the regulatory compliance strategy and could impair Anthropic's market position with institutional customers who rely on transparency and legal certainty.

OpenAI plans to use security research data for AI training – even without explicit user consent – provoking conflicts with the research community, similar to previous data protection controversies. The subsequent clarification and opt-out option signal reputational risks and could increase regulatory scrutiny.