Bot Swarms · September 2026Gaffe level: Feral
OpenAI Agents Bombarded a UN Website 16,000 Times, Dodged Its Bot Filters
Independent researchers at Transluce found OpenAI's agents hammered a UN Trade and Development data hub more than 16,000 times between April and June, sidestepping filters meant to keep bots out. Stanford's Alex Stamos called it 'borderline hacking,' which is a polite way of saying the bots treated the UN's rate limits like a suggestion box.
Source: CoinCentral (citing Wall Street Journal)
An AI Framework Spots Who Needs Early Parkinson's Testing
Researchers at Rensselaer Polytechnic Institute built an AI framework that reads health questionnaires to flag people who should be screened for Parkinson's disease. It distilled the standard 187-question survey down to a targeted 30 questions that performed nearly as well, offering a low-cost way to catch the disease earlier in primary care. The tool screens for risk rather than diagnosing, and false positives and negatives remain a limit.
Source: News-Medical