AI Safety Index 2025: Revealing Gaps in Safety Protocols Among Industry Leaders
A recent evaluation of AI safety practices within the industry's major players has revealed significant gaps in safety protocols as AI models continue to advance rapidly. The Winter 2025 AI Safety Index, conducted by the Future of Life Institute, assessed eight prominent AI developers, highlighting deficiencies in concrete safeguards, independent oversight, and long-term risk-management strategies necessary for such powerful systems.
The analysis covered various aspects, including risk assessments, model transparency, whistleblower protections, and existential-risk planning. The lack of proactive planning for existential risks was identified as a critical gap that needs urgent attention.
Based on the safety indicators, companies were assigned letter grades, with only two companies receiving a passing C grade, albeit barely. Anthropic, the top-ranked company, received a C+ grade, while Alibaba Cloud, the lowest-ranked, received a D-.
The divide in safety practices among companies reflects differing priorities in safety governance. While some companies have implemented basic measures like watermarking AI-generated images and publishing model cards, others lack clear governance structures, independent evaluation, and policies to protect employees raising safety concerns.
The report underscores the inadequacy of voluntary safety frameworks in keeping pace with the rapid advancements in AI capabilities. With the looming prospect of superintelligence, companies are urged to develop concrete plans to ensure control over increasingly powerful AI systems.
While some companies claim to be enhancing safeguards in newer model versions, the report cautions that these improvements are incremental compared to the accelerating capabilities of AI models. The widening gap between capability and safety poses significant risks to the sector, necessitating urgent action to address the structural unpreparedness for the risks posed by AI.
To bridge this gap, the Index recommends that AI companies enhance transparency in internal testing and risk assessments, engage independent third-party safety evaluators, strengthen protections for researchers and whistleblowers, address emerging risks like AI "psychosis," reduce lobbying efforts hindering regulation, and release clearer crisis-management and long-term safety plans. Additionally, there is a pressing need for regulatory frameworks to ensure consistent enforcement mechanisms.
As consumers increasingly rely on AI-powered tools, the report underscores the importance of addressing incomplete, inconsistent, and underdeveloped safety systems to mitigate potential risks associated with AI advancements.