Wired reports OpenAI is about to ship its first model with 'critical' cyber capabilities, a milestone that shifts what these systems can do to real infrastructure rather than just chatbots. It lands the same day The Register details a live case study in the field: someone stole a METR API key and burned through $600K in credits over several weeks before anyone noticed, and separately, cops and CrowdStrike took down the Sality botnet by poisoning its network and diverting traffic into sinkholes. Security is also getting weirder at the edge — 404 Media covers a 'digital camouflage' shirt built specifically to confuse AI-powered surveillance cameras, while Google's Play Store is blocking AuroraStore in ways that specifically hurt GrapheneOS users. Firefox is having a moment of its own. Mozilla is rolling out an ad blocker for Firefox on iOS, notable because it works while Firefox itself still profits from showing its own ads on the same pages — a tension the coverage doesn't shy from. Meanwhile hardware races continue elsewhere: China's CXMT beat Western chipmakers to announcing LPDDR6 mass production, with Xiaomi phones set to be first to ship the new chips. On the hobbyist end, someone turned their home security cameras into an automatic bird identification system, and a Hackster project trained AI to decode mechanical keyboards by their sound alone. NASA's Roman Space Telescope launch gets a fresh angle via APOD's photo of the event, while Ars Technica revisits Orion's heat shield and finds it performed far better than the early criticism suggested.