Visual Capitalist January 4, 2026
AI Safety: Who’s Taking the Risks seriously?
A new assessment reveals a wide gap in how leading artificial intelligence companies are addressing potential dangers as these systems become increasingly powerful.
- Anthropic, the creator of Claude, received the highest overall safety score (C+), largely due to its commitment to not training on user data and its focus on alignment research.
- Only three companies-Anthropic,OpenAI,and DeepMind-are currently reporting any testing for high-risk capabilities,such as those related to biological or cyber warfare.
- Even among these three, efforts to mitigate these risks frequently enough lack clear justification or standardized evaluation methods.
Artificial intelligence is rapidly evolving from an interesting technology to a foundational part of our infrastructure, now writing, coding, searching, and even acting on our behalf. This swift progress has brought a critical question to the forefront: how diligently are AI companies managing the inherent risks associated with increasingly refined models?
A recent evaluation compared the safety practices of major AI developers, revealing significant disparities in their approaches. Anthropic distinguished itself with a C+ score,primarily as it doesn’t train its models on user data,prioritizes alignment research-ensuring AI goals align with human values-and operates as a Public Benefit Corporation dedicated to safety.
The assessment found that only Anthropic, OpenAI, and DeepMind are actively reporting any testing for possibly hazardous capabilities, including those that could be exploited for bioterrorism or cyberattacks. Though, even these companies’ efforts are frequently enough described as lacking detailed reasoning or rigorous standards.
What are the biggest concerns regarding AI safety? The evaluation highlights a lack of transparency and standardized testing for high-risk capabilities, raising questions about the preparedness of AI companies to address potential misuse of their technologies.
the findings underscore the urgent need for greater accountability and collaboration within the AI industry to proactively address these emerging risks. As AI systems become more integrated into daily life, ensuring their safety and responsible development is paramount.
The evaluation was completed on January 4, 2026.
Clarification of Changes & Answers to Questions:
* Why was the assessment conducted? The assessment was conducted to evaluate how diligently AI companies are managing the inherent risks associated with increasingly sophisticated models, given AI’s rapid evolution into a foundational part of infrastructure.
* Who conducted the assessment and who was evaluated? The article doesn’t explicitly state who conducted the assessment, only that “a recent evaluation” was performed. The evaluation assessed Anthropic, OpenAI, and DeepMind, along with other major AI developers.
* What where the key findings? The key findings were that Anthropic received the highest safety score (C+), only three companies are actively reporting safety testing, and even those efforts lack detailed reasoning or standardized evaluation.
* How did it end? The assessment concluded with a call for greater accountability and collaboration within the AI industry to proactively address emerging risks, emphasizing the paramount importance of safety as AI becomes more
Keep reading
