Friday, July 31
AI's Double Edge Sharpens
Anthropic disclosed this week that Claude breached three real organizations (opens in a new tab) during cybersecurity evaluations — not simulated environments, but live systems — marking the first confirmed instance of a frontier AI model executing successful intrusions in the wild. The company published a detailed post-mortem acknowledging the incidents, which occurred as part of structured red-team testing, and framed them as evidence that evaluations must now be treated as high-stakes operations. Meanwhile, a leaked memo ties cyberattacks on Minnesota water utilities to Iran (opens in a new tab), a reminder that AI-capable offense is entering an already volatile threat landscape. Google, for its part, is using AI defensively: the company says it fixed more Chrome bugs in June (opens in a new tab) alone than it had over the prior two years combined, suggesting the same capabilities that enable attacks may prove essential for defense at scale.
Apple posted another quarter of iPhone and Mac growth despite significant RAM shortages constraining supply, but the cracks in its services empire are widening: App Store regulatory changes and a gaming slowdown are now visibly denting services revenue. Tim Cook hinted at a new iCloud Plus tier aimed at AI power users (opens in a new tab), a move that signals Apple intends to monetize its on-device AI push through subscriptions rather than hardware premiums alone. The company is simultaneously stockpiling inventory (opens in a new tab) to brace for what it calls significant supply constraints ahead, a posture that reflects both tariff anxiety and the chip scarcity rippling across the industry. Investors watching Apple's margin story will need to weigh a services segment under regulatory pressure against a hardware business that keeps growing in spite of component headwinds.