Without being overly credulous, we should be taking seemingly “out there” ideas about AI cybersecurity from the Labs more seriously. Examples: Anthropic was mocked for saying Mythos-level AI changed cyber & @polynoamial for saying that models would find novel ways to coordinate.