The Dark Factory Is Real: OpenAI Ships Code Nobody Reviews, While Anthropic Warns of "First Clear and Present Danger"

AI-written code deployed without human review is moving from experiment to default. OpenAI's Ryan Lopopolo describes the "Dark Factory" — a million lines of code and a billion tokens a day running with zero human reviewers — while Sam Altman announces 3 million weekly Codex users. At the same time, Anthropic unveils Project Glasswing and Claude Mythos Preview, a cybersecurity model so capable at finding exploits that Anthropic withheld the weights. CEO Dario Amodei called cyber "the first clear and present danger" from advanced AI. Meanwhile the model landscape splits: Meta ships its first closed frontier model (Muse Spark) from Superintelligence Labs, while Zhipu AI (GLM-5.1) and Google DeepMind (Gemma 4) push maxima ...