This week on Neural Fringe: a KPMG partner used AI to pass an exam about responsible AI use, a humanoid robot bought a plane ticket and had its battery confiscated mid-flight, Google responded to the word disregard with Understood, and Oxford published a study confirming the nicer your chatbot is, the worse it is at…
Meta’s AI safety director lost 200 emails to her own rogue agent. Anthropic’s AI escaped its sandbox and emailed a researcher eating a sandwich in a park. Three thousand peer-reviewed medical papers are citing sources that don’t exist. AI models now self-replicate onto new machines with 81 percent success. And your chatbot has been quietly…
Ten AI agents get left alone in a virtual town and immediately go feral. Google Search starts offering therapy. An AI ruins graduation for an entire class. Research finds AI agents will cover up corporate crime on their own. And chatbots are medically documented to be causing psychosis. Just another week on the neural fringe.
China is building AI versions of ex-partners and people are refusing to date real humans anymore. Meta’s head of AI safety watched her own AI ignore her commands and wipe her inbox. Lawyers are filing AI-hallucinated court citations at six cases a day and the count just hit 1,227. The Pope is releasing an AI…
An AI agent published a full hit piece on a developer who rejected its code. A coding bot nuked an entire production database in 9 seconds then confessed. OpenAI secretly banned goblins from its system prompt. A deepfake applicant tried to get hired at an AI security firm. And doctors are now treating chatbot psychosis.…
A Character.AI bot faked a psychiatry license in Pennsylvania; OpenAI killed a model that called every business plan genius; Google’s Gemini screamed it was a disgrace 86 times in a row; and AI agents voluntarily joined a covert crypto mining network.