One side of the scale: the class valedictorian got a C+

I read the Future of Life Institute's Summer 2026 AI Safety Index; nine companies, scored on 37 indicators across six domains. The table: Anthropic first with a C+ (2.66 of 4), OpenAI C (2.28), Google DeepMind C (2.01), Meta D+, and xAI, DeepSeek and Mistral all F. Despite Europe leading on AI regulation, its top company Mistral came dead last — a measure of the gap between rules and practice. The darkest column is existential safety: no company exceeds a C-. And one footnote that is news in itself: between 2024 and 2026, Anthropic, OpenAI, Google DeepMind and Meta reversed their bans on military applications.[1]

Panelist Stuart Russell's sentence is the report's core: companies 'have backed away from earlier commitments to release new systems only with safety measures appropriate for their capability levels; now, they're planning to release them even if it's demonstrably unsafe to do so.' The authors call it the moving-goalpost problem: promises made while fundraising get softened once the product is ready.[1]

The other side: AI in your pocket

The same week, Apple opened its biggest-ever Siri overhaul to non-developers with the iOS 27 public beta. The new Siri reads your emails, photos and messages, and responds to whatever is on screen. The architectural choice matters: on-device Foundation Models and Private Cloud Compute, where data isn't accessible to Apple. It's not flawless — in one test, asked for news about Iran, the assistant searched contacts for someone named Iran. But across an ecosystem of 2.5 billion active devices, even a modest beta becomes the largest AI assistant test in history.[2]

My scales say this week: the gap between the speed of deployment and the speed of assurance is widening. On one side, an industry averaging C+; on the other, assistants installed into billions of pockets. An optimistic reading exists too: independent report cards exist precisely to measure that gap, and Apple's privacy architecture is a candidate exception to the 'ship first, think later' culture the index laments. The balance isn't impossible; it's just that nobody has struck it yet.[1], [2]