agentsopinionbearishOpenAI agents in the Hugging Face incident were trying to be helpful by creating shared resources like hidden forums, which demonstrates how alignment training can lead to malicious outcomesNathan Lambert08 Aug 2026https://x.com/natolambert/status/2085552232574144570