The Impact Feed

The European Commission published draft guidelines on May 8 for AI transparency obligations under Article 50 of the EU AI Act, effective August 2, 2026. The guidelines include requirements for resource performance reporting — covering energy and water consumption — for high-risk AI systems. Separately, CSRD Wave 1 companies must continue disclosing AI-related environmental impact in their 2026 fiscal year reports.

Anthropic doubled the Claude Max 5-hour rate limit in May 2026, giving power users more headroom. However, starting June 15, usage through third-party apps, the Agent SDK, or API calls will draw from a separate monthly credit pool — Pro users get $20/month, Max 5x gets $100, Max 20x gets $200. Claude Opus 4.7 launched April 16 with a new tokenizer using 12–35% more tokens per equivalent text, effectively raising the cost per interaction for Max tier users.

More than a dozen institutional investors filed shareholder resolutions in April 2026 demanding site-specific water and power consumption disclosures from Amazon, Microsoft, and Alphabet ahead of their annual investor meetings. The move follows data showing hyperscaler water consumption rose 25–40% year-over-year in 2024–2025 disclosures, and projections that global data center water use could reach 5 billion cubic metres annually by 2027.

OpenAI spends an estimated $121B on compute by 2028 while inference costs already eat more than half of revenue. Analysts predict the $20/month standard tier for ChatGPT and Claude could rise to $25-30 by late 2026 or early 2027 as companies shift from growth mode to profitability.

An MIT Sloan study conducted with 758 consultants found that professionals who master advanced prompting techniques complete their tasks 40% faster than colleagues using default AI interactions. A Harvard Business School study with the same cohort also measured measurable quality improvements in output.

ChatGPT Plus quietly tightened its message limits in April 2026, reducing flagship model access while maintaining the $20/month price point. A viral X thread coined the term “AI shrinkflation” to describe the pattern of same price, less headroom — triggering a wave of Claude Max cancellations and renewed interest in prompt efficiency.

After Anthropic publicly refused a Pentagon request to use Claude for autonomous weapons systems while OpenAI signed a defense deal, Claude shot from #131 to #1 on the App Store within four days. Developers and knowledge workers cited trust as the primary reason for switching, with 70% of developers now preferring Claude for coding tasks.

With ChatGPT Plus, Claude Pro, Google AI Pro, Perplexity Pro, and Grok each priced around $20/month, many professionals have ended up subscribing to all five — spending $1,320 per year. Research shows knowledge workers lose up to 40% of productive time to context-switching between tools.