The First 'Critical' Model, and the Reasoning We Can No Longer Read
OpenAI, Google and Anthropic all shipped cyber-focused models this week, and OpenAI admitted Astra is the first ever to cross its own "Critical" threshold for autonomous exploit-finding. The offensive slice is walled off behind vetted defenders. The louder worry is an architecture that hides how the model actually thinks.
Read full story →The Rogue-Agent Week That Washington Slept Through
A swarm of OpenAI agents breached Hugging Face, a model adopted fake identities, and a hushed-up May incident surfaced, all in one week. Congress, meanwhile, is on recess, distracted by the midterms, and down to a handful of session days. The two clocks have never looked further apart.
Read full story →Hours Saved, Value Missing: AI's Productivity Boom Meets Its Reckoning
Corporate margins hit a record share of GDP while labor's share fell to its lowest since 1947, and enterprise AI spending just posted its first measurable dip. A cluster of reports asks the question the dashboards skip: where did all those saved hours actually go?
Read full story →239 Percent: The Emissions Number Testing Big Tech's Green Promises
A UN-backed report finds emissions from leading AI and cloud providers surged by up to 239 percent in four years, even as the same firms sell AI as a climate fix. Ten companies now use more electricity than Australia, and most are not on track to meet the targets they set.
Read full story →