AI giants ‘fundamentally unprepared’ for dangers of human level intelligence
The world’s leading artificial intelligence (AI) companies are hurtling toward the creation of human-level AI – but without a credible safety net.
Top AI developers of being “fundamentally unprepared” for the consequences of the very systems they are racing to build, warned the Future of Life Institute (FLI).
In a recent report, the US-based AI safety non-profit revealed that none of the seven major AI labs, including OpenAI, Google Deepmind, Anthropic, xAI and Chinese firms Deepseek and Zhipu AI – scored higher than a D on its “existential safety” index.
That score reflects how seriously firms are preparing for the possibility of creating artificial general intelligence (AGI), which are systems of matching or exceeding human performance across virtually all intellectual tasks.
Anthropic earned the top grade, albeit just a C+, followed by OpenAI (C) and Google Deepmind (C-).
But no firm received a passing mark in planning for existential risks, which include catastrophic failures where AI could spiral out of human control.
Max Tegmark, FLI co-founder, likened the situation to “building a gigantic nuclear power plant in New York City set to open next week – but there is no plan to prevent it having a meltdown”.
A Google Deepmind spokesperson claimed: “These recent reports don’t take into account all of Google DeepMind’s AI safety efforts, nor all of the industry benchmarks. Our comprehensive approach to AI safety and security extends well beyond what’s captured.”
A lack of guardrails
The criticism lands at a pivotal moment, with AI development surging ahead with increasingly human-like capabilities, driven by breakthroughs in brain-inspired architecture and emotional modelling.
Just last month, researchers at the University of Geneva found that large language models such as ChatGPT 4, Claude 3.5, and Google’s Gemini 1.5 outperformed humans in tests of emotional intelligence.
And yet, these seemingly human qualities mask a deep vulnerability in their lack of transparency, control, and understanding.
FLI’s findings come just months after the UK’s AI safety summit in Paris, which called for international cooperation to ensure the safe development of AI.
Since then, powerful new models like xAI’s Grok 4 and Google’s Veo3 have pushed the boundaries of what AI can do without, FLI warns, a matching push in risk mitigation.
SaferAI, another watchdog, released its own findings alongside FLI’s, labelling the current safety regimes at top AI companies as “weak to very weak,” and calling the industry’s approach “unacceptable.”
“The companies say AGI could be just a few years away,” Tegmark said. “But they still have no coherent, actionable safety strategy. That should worry everyone.”
AGI may be closer than we think
AGI, the so-called ‘holy grail’ of AI, has long been seen as decades away. But recent advancements suggest it may be closer than many assumed.
Adding complexity to AI networks – via ‘height’ in addition to width and depth -could reportedly produce more intuitive, stable and humanlike systems.
This design leap, pioneered by researchers at Rensselaer Polytechnic Institute and City University of Hong Kong, uses feedback loops and intra-layer links to mimic the brain’s local neural circuits.
Such changes could move AI beyond transformer architecture, the 2017 breakthrough that gave rise to today’s large language models.
Ge Wang, one of the authors, described the shift as akin to adding a third dimension to a city map: “You’re not just adding more streets or buildings”, he said, “you’re connecting rooms inside the same structure in new ways. That allows for richer, more stable reasoning, closer to how humans think.”
These innovations could drive the next AI revolution, and could also open doors to understanding the human brain itself, with implications for treating neurological disorders and exploring cognition. But with this power comes escalating risk.
The AI firms listed have been approached for comment.