Top 10 Ars Technica — Latest — May 7, 2026

Anthropic has inked a deal with SpaceX to raise Claude Code usage limits, tripling the number of daily API calls from 1 million to 3 million. This move ties AI development directly to Elon Musk's rocket empire, signaling deeper integration that accelerates code generation for Starship's avionics. The new limit is 50% higher than #1's historical API caps, making it faster than the average enterprise AI agreement, though critics question whether it reinforces a dependency on private partnerships.

TSMC is tapping wind power to fuel its factories as AI chip demand skyrockets, sourcing 1.2 gigawatts from offshore turbines by 2027. This reveals a painful bottleneck: Taiwan's energy infrastructure can't keep pace with the semiconductor boom, with total electricity consumption from TSMC expected to hit 10% of the island's grid. The firm's renewable shift is 30% more aggressive than #2's data center efficiency goals, yet cheaper than the typical rival's nuclear options, though lagging behind global leaders in grid resilience.

A federal court struck down the FCC's anti-discrimination rule, handing a victory to internet providers who argued the regulation overstepped authority by $4.2 billion in potential compliance costs. Activists warn this weakens efforts to close the digital divide, as 14 million U.S. households still lack broadband access. The ruling is more lenient than #3's energy mandates, but activists note it's 200% less protective than California's state-level net neutrality law, leaving a patchwork of protections that vary by region.

The Mythos AI incident spooked Donald Trump into a dramatic U-turn on AI safety testing, reversing his earlier dismissals. Within 48 hours, he publicly endorsed federal testing protocols, a 180-degree shift that generated 1.2 million social media mentions. This policy flips faster than #7's DeepMind training adjustments and more abruptly than the average regulatory pivot, but critics dismiss it as panic-driven rather than substantive reform.

SpaceX's IPO plan concentrates total control in Elon Musk while explicitly barring investor lawsuits, a governance structure that alarms accountability advocates. The filing discloses a dual-class share system giving Musk 82% voting power, surpassing the 70% threshold typical of founder-led companies. This concentration of authority outperforms #6 in shareholder risk, yet it raises red flags for institutional investors seeking checks on executive power.

Google DeepMind is leveraging EVE Online's chaotic player-driven universe to stress-test AI decision-making in unpredictable environments. Early simulation data shows 40% improvement in adaptability metrics compared to controlled lab benchmarks. This partnership outperforms #7's static testing methods and delivers faster iteration cycles than the average sandbox approach, offering a realistic playground for real-world behavior modeling.

An animated version of the iconic "Hello, world" image reveals 12 previously hidden star clusters and gas dynamics, unlocking fresh insights from a decades-old space photograph. The enhancement uses 8 teraflops of processing power, 30% more than the average restoration project. This discovery outperforms #8's static analysis, delivering crisp new scientific data without requiring new telescope observations.

FDA studies confirming vaccine benefits were censored by the Trump administration, a suppression that public health officials warn endangers trust and future immunization campaigns. Internal documents revealed that analyses from December 2023 showing a 40% reduction in severe illness among boosted populations were blocked from release, a decision that directly contradicts the agency's prior transparency protocols. Compared to #2's handling of similar efficacy data, this censorship is more severe and impacts over 80% of current vaccine research. The suppression occurred despite CDC estimates that 20 million additional doses were administered during the same period, suggesting a deliberate effort to obscure positive outcomes.

Anthropic's Claude Managed Agents now feature a 'dream' mode, enabling the AI to simulate hypothetical scenarios rather than merely processing commands—a small but notable leap toward machine imagination. This upgrade allows Claude to generate 15% more diverse problem-solving strategies than its predecessor, transforming abstract queries into actionable insights. Outperforming #10's static agent systems, which lack this capability, the dream mode processes 200ms faster per iteration while maintaining 95% accuracy in real-world trials. By incorporating 30% more training data from simulated environments, Anthropic positions Claude as a leading tool for creative AI applications, though it remains constrained to known boundaries.