AI spending is not plateauing. 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 a single point: costs are rising, not falling. Levie cited Box's own token consumption as proof. Early AI tasks consumed 5,000 to 20,000 tokens. Today, Box agents run tasks at 1 million to 5 million tokens. Moore's Law is not keeping up because companies keep routing harder problems through these systems, not cheaper ones.
Two sharper findings cut deeper than the spending numbers. First, Alex Stamos, former Meta security chief and current Corridor CPO, argued the real AI cybersecurity inflection already happened, quietly, with the Opus 4 and GPT-5 series models. Those gave human engineers what Stamos called Olympic-level capabilities at finding vulnerabilities. Mythos currently leads public models in bug detection. Second, Levie pointed out that the U.S. government's current posture of reviewing and approving individual models is functionally identical to what the 2023 six-month AI pause petition demanded. The pause crowd lost the argument and won the policy.
The original piece also covers Anthropic's Fable, which Krieger described as reducing the back-and-forth loop between humans and AI agents, and includes direct quotes from Brockman on the state of frontier development. The summit compressed a month of industry debate into hours, and the full transcript details where each speaker diverged, which is where the real signal lives.
[READ ORIGINAL →]