The Winter edition of AI Safety Index from Future of Life Institute shows that xAI, Z.ai and Meta have increased transparency regarding their risk assessment procedures, mirroring the practices of Anthropic, OpenAI, and Google DeepMind.
The report arrives at a moment of intensified public scrutiny around the societal impact of increasingly capable AI systems, particularly following several high-profile incidents where advanced chatbots were linked to cases of self-harm and suicide.
As noted, more companies have conducted internal and external evaluations of AI risks. The assessments indicate a narrow risk scope, limited validity, and external reviews that are not entirely independent.
According to the report, despite the absence of Chinese companies in the Top 3, reviewers identified and noted a number of effective safety practices resulting from domestic regulatory mandates. China’s domestic regulations, including binding rules for content labeling, incident reporting, and voluntary AI risk-management standards, establish a higher minimum safety accountability baseline for Chinese firms relative to Western counterparts.
The safety practices demonstrated by companies generally do not align with the benchmarks defined in emerging standards, such as the EU AI Code of Practice.
These reports show an imbalance where safety commitments are disproportionate to capability ambition. While even top companies lack the necessary concrete safeguards, independent oversight, and long-term risk protocols, the majority of the industry fails basic transparency and governance. This gap suggests the sector is structurally unprepared for the risks it is actively introducing.
