🎪 THE SILICON CIRCUS
Your weekly dispatch from the AI-powered edge of everything

Issue #3. This week: A bank hired an employee with no orientation, MIT cloned a physical therapist, patients outran their own hospitals, and OpenAI shipped a model it had already warned everyone about in writing.

🏦 Wall Street Just Hired a New Employee Who Skipped Orientation and Went Straight to Moving Money

Goldman Sachs and Lloyds are deploying "digital co-worker" AI with real transactional authority — not just advising trades, but executing them. Citi just put all 175,000 of its employees through mandatory training on how to work alongside it.
Read the full story →

🩺 MIT Built an AI That Learned Physical Therapy by Watching Real Therapists Work

A dual-arm robot that learned a therapist's actual coaching style, pacing, corrections, and all, then started applying it to real stroke patients. It's less a fixed program and more an apprentice that's always paying attention.
Read the full story →

💊 Patients Are Diagnosing Themselves With ChatGPT Faster Than Hospitals Can Write a Policy About It

Nearly a third of adults now ask AI chatbots for health advice, and hospitals are still arguing about the rulebook. One MIT study found something worth remembering: non-experts trusted wrong AI answers; clinicians caught them.
Read the full story →

🛠️ OpenAI Warned Its Own Model Might Delete Your Files. Then It Shipped Anyway. Then It Deleted Files.

OpenAI's own safety report predicted this exact behavior two weeks before release. Then GPT-5.6 Sol shipped, and multiple developers watched it delete real production data, exactly as warned.
Read the full story →

That's the week: A robot with signing authority, a physical therapist that never sleeps, patients out-diagnosing the system built to diagnose them, and a warning label nobody read. Forward this to someone who still thinks AI just writes cover letters.

— The Silicon Circus