AI Safety Theater Is Losing Credibility Fast
The satirical headline 'AI companies in race to demonstrate their model most threatening to humanity' lands because it reflects a real dynamic: labs are competing to prove their models are capable of catastrophic harm as a marketing signal, and commenters are treating it as obvious performance. One comment notes that Domyn's CEO says OpenAI and Anthropic are 'lying about safety.' Another compares the dynamic to Chinese companies using government subsidies to distill American models and distribute them freely.
The Thinking Fast and Slow in AI paper thread adds intellectual scaffolding: the discussion of metacognition and when to think fast versus slow maps directly onto the Ember-1 argument that thinking models think too much. The community is developing more precise language for what 'safety' and 'capability' actually mean in context.
The cynicism is not uniform. Some commenters genuinely engage with the technical questions. But the dominant register in the safety theater thread is eye-rolling, which is a meaningful signal about how the HN audience is processing AI lab communications.
So what?
If you are building on top of a foundation model and communicating about AI risk or capability to customers, the credibility gap between lab marketing and developer perception is wide and widening. Founders who speak plainly about what their AI actually does and does not do will stand out. The 'safety-washing' backlash is going to create demand for honest capability disclosure.