A report by the Future of Life Institute (FLI) highlights concerns about AI companies’ preparedness for risks tied to human-level AI.

The index evaluated seven AI developers, including OpenAI, Google DeepMind, Anthropic, Meta, xAI, Zhipu AI, and DeepSeek, across six categories, such as “current harms” and “existential safety.”

None of the seven companies evaluated received higher than a D grade in “existential safety planning” on FLI’s AI safety index.

Anthropic scored the highest with a C+, followed by OpenAI (C) and Google DeepMind (C-). The index assessed areas like “current harms” and “existential safety.”

Max Tegmark, FLI co-founder, likened the situation to building a nuclear plant without a safety plan.

He warned that AGI development is progressing faster than expected, with some companies predicting it could arrive in a few years.

The report was reviewed by experts including Stuart Russell and Sneha Revanur.

🔗 Source: The Guardian

🧠 Food for thought

1️⃣ AI development timelines have compressed dramatically with safety lagging behind

AI has evolved from theoretical foundations laid by Alan Turing in 1950 to today’s advanced systems approaching general capabilities, compressing development timelines significantly1.

While early AI development progressed gradually through distinct phases, such as symbolic AI (1956-1974), expert systems (1980-1987), and neural networks (1993-2009), recent advances have accelerated dramatically1.

Most notably, expert predictions for AGI development have shortened considerably, with some researchers now suggesting AGI could arrive as early as the 2030s, while companies like OpenAI claim even shorter timeframes23.

This acceleration contrasts sharply with safety planning efforts, with the FLI report revealing that no company scored higher than a D for “existential safety planning” despite their ambitious development goals.

When the AI100 Project launched in 2014 to study AI’s societal impact over a century, few anticipated how quickly capabilities would advance, creating a widening gap between technological progress and safety frameworks4.

2️⃣ Current AI safety approaches lack the rigor of established engineering fields

Recent OpenAI developments