AI company efforts ranked in safety report card for protecting humanity

A recent assessment questions whether artificial intelligence companies are adequately protecting humans from AI-related risks. The findings suggest that the industry has significant gaps in safety practices as AI becomes increasingly embedded in daily life, from mental health chatbots to potential cyberattack tools.

The AI Safety Index, released Wednesday by the Silicon Valley nonprofit Future of Life Institute, aims to guide AI development toward safer practices and reduce existential risks to humanity. The report highlights that current incentives for AI firms do not prioritize human safety.

"This industry is largely unregulated in the U.S., creating a competitive environment where safety is often neglected," explained Max Tegmark, president of the institute and MIT professor.

The highest grades awarded in the report were only C+, received by OpenAI, developer of ChatGPT, and Anthropic, known for its Claude chatbot. Google DeepMind received a C, while Meta and Elon Musks xAI were given a D. Chinese companies Z.ai and DeepSeek also earned a D, and Alibaba Cloud scored the lowest with a D-.

Grades were based on 35 indicators across six categories, including existential safety, risk assessment, and transparency. Evaluators considered public documentation and company responses to surveys, scoring the firms through a panel of eight AI experts including academics and organizational leaders. All companies scored below average in existential safety, which examines internal monitoring, control mechanisms, and long-term safety planning.

The report stated, "While companies pursue artificial general intelligence and superintelligence, none have shown a credible plan to prevent catastrophic misuse or loss of control."

OpenAI emphasized its commitment to safety: "We invest in frontier safety research, implement safeguards, rigorously test models internally and with independent experts, and share safety frameworks to help establish industry standards."

Google DeepMind similarly stated that it follows a "science-led approach to AI safety," with a Frontier Safety Framework to identify and mitigate risks from advanced models.

The report criticized xAI and Meta for lacking monitoring and control commitments despite having risk frameworks. DeepSeek, Z.ai, and Alibaba Cloud also did not provide public documentation on existential safety strategies. Several companies did not respond to comment requests.

Tegmark warned that insufficient regulation could enable misuse of AI for bioweapons, manipulation, or political destabilization, but noted that establishing binding safety standards could address these risks effectively.

Government oversight efforts exist but face resistance from tech lobbying groups concerned about innovation slowdown. Californias SB 53, signed by Gov. Gavin Newsom, requires AI companies to report safety protocols and incidents such as cyberattacks. Tegmark called it progress but emphasized the need for broader measures.

Analyst Rob Enderle described the AI Safety Index as an innovative approach but expressed concerns about the effectiveness and enforceability of U.S. regulations.

Analysis: AI Safety Practices Remain Inadequate

The recent AI Safety Index highlights a persistent gap between the rapid development of artificial intelligence and the measures companies take to protect humans from potential risks. Even leading firms such as OpenAI and Anthropic received only C+ grades, signaling that safety practices are still below the level necessary for robust risk mitigation.

All evaluated companies scored poorly in existential safety, indicating weak internal monitoring, insufficient control mechanisms, and a lack of long-term safety planning. The report underscores that while firms pursue advanced AI, none have credible strategies to prevent catastrophic misuse or loss of control.

Regulatory frameworks remain limited in the U.S., leaving companies to self-regulate in a competitive environment. California’s SB 53 represents a step forward by mandating disclosure of safety protocols and incidents, but broader, enforceable standards are needed to prevent potential misuse in areas like cyberattacks, bioweapons, or political manipulation.

Industry statements, including those from OpenAI and Google DeepMind, emphasize commitment to safety research and frameworks, yet the overall low scores suggest these efforts are not sufficient to ensure consistent, industry-wide protection. Without stronger oversight and incentives, the risk of unsafe AI deployment is likely to grow as these technologies become more embedded in daily life.

Sources:

• A safety report card ranks AI company efforts to protect humanity