The world’s leading AI companies are suffering from a ‘deeply disturbing’ lack of safety planning in the race to create human-level artificial intelligence, according to prominent AI safety researchers.

Publishing its latest AI Safety Index, which rates the safety and security risks of top-level AI firms, the US non-profit Future of Life Institute (FLI) found that despite claims they will achieve artificial general intelligence within the next decade, none of the leading AI developers scored an overall safety grade higher than a C+.

According to the FLI ranking, among the worst performers are Chinese-owned firms Zhipu AI and the by now globally recognised AI disruptor, DeepSeek.

Both scored the lowest possible safety grade, F, with the report finding major issues across all categories, including risk assessments, safety frameworks, information sharing, and, crucially, existential safety, which examined companies’ preparedness for managing extreme risks from future AI systems that match or exceed human capabilities.

Also performing extremely poorly in this area, Meta and OpenAI were found lacking in existential safety strategies, internal monitoring and controls, technical safety research, and support for external research.

The FLI made special mention of what it described as OpenAI’s ‘deteriorating safety culture’ after the firm lost most of its researchers in this area over the last year, which led to a failing grade for the company, whose mission is to see AGI ‘empower humanity to maximally flourish in the universe’.

None of the seven firms the FLI benchmarked managed to score above a D in existential safety, though Anthorpic was praised for its detailed and transparent research regime, while Google DeepMind’s safety documentation was described as “well thought out, with a serious commitment to monitoring”.  

Overall, however, the Index raises serious questions about the guardrails the firm’s leading the AGI charge have in place, with one reviewer warning that “literally none of the companies has anything like a coherent, actionable plan”. 

That could be a bitter pill to swallow for some of these AI firms’ researchers, however. Posting to X earlier this week, OpenAI safety researcher and professor of Computer Science at Harvard, Boaz Barak, said that the way xAI handles safety is “completely irresponsible”.

Barak said that xAI’s Grok has “no system card” and “no information about any safety or dangerous capability evals”, something he said is standard across competitors like OpenAI, Anthropic, Meta and Google DeepMind.

Recommended reading

• Meta Unveils Billion-Dollar AI Data Centre Push
• ChatGPT Search Inches Toward Strict EU Oversight as Users Surge
• Beyond DeepSeek: 3 Critical Questions For The Future of AI

However, going by the FLI’s Index, both xAI and OpenAI received failing grades in existential safety, while OpenAI scored just a single grade higher than Elon Musk’s AI outfit for overall safety.    

“These findings reveal an unregulated industry where competitive pressures and technological ambition far outpace safety infrastructure and norms,” cautioned the report.

“This imbalance becomes more dangerous as companies pursue their stated goal of achieving artificial general intelligence that matches or exceeds human capabilities within the decade. 

“The industry is fundamentally unprepared for its own stated goals.”