AI spending is not slowing down. At the Big Technology AI Summit in San Francisco last week, speakers including OpenAI President Greg Brockman, Anthropic Labs Lead Mike Krieger, and Box CEO Aaron Levie converged on one point: costs are rising because capability is outpacing efficiency gains. Levie put a number on it: Box tasks that once consumed 5,000 to 20,000 tokens now routinely hit 1 million to 5 million tokens per agent execution. Moore's Law is not keeping up.
The cybersecurity signal buried in this summit deserves more attention than it got. Alex Stamos, former Meta security chief and current Corridor CPO, argued the real AI cyber threshold was crossed last year with the Opus 4 and GPT-5 series models, not with any recent export control announcement. His framing: human engineers suddenly running Olympic times at a high school track meet. Stamos named Mythos as the current top public model for bug-finding. The White House is reviewing new models before release, which Levie noted is functionally the regulatory pause that the 2023 open-letter signatories wanted, delivered through a different mechanism entirely.
The full piece covers Krieger on how Fable reduces iteration cycles, Brockman's read on where frontier development goes next, and the specific mechanics behind what the summit called the 'tokenmaxxing reckoning.' The argument about why AI costs behave differently than every prior computing cost curve is worth reading in Levie's own words.
[READ ORIGINAL →]