AI companies retreat from safety pledges as industry-wide rankings reveal troubling decline

AI Safety Rankings Expose Industry-Wide Shortcomings

The Future of Life Institute released its latest AI Safety Index on Tuesday, revealing a stark reality: not a single AI company received an A grade in any category. Anthropic, despite building its brand on AI safety, secured the top position with a modest C+ overall score, while OpenAI slipped from C+ to C and Google DeepMind ranked third with a C grade 1

1

2

2

. The semiannual assessment, conducted by seven researchers and governance experts, evaluated nine leading AI companies across six categories: risk assessment, current harms, safety frameworks, existential safety, governance and accountability, and information sharing 3

3

.

Meta showed improvement, climbing from sixth to fourth place with a D+ grade, while xAI dropped three spots to seventh with an F grade 2

2

. The failing grades extended to China's DeepSeek and France's Mistral, which placed last among the nine companies evaluated. Max Tegmark, MIT professor and Future of Life Institute president, noted the geographic spread of poor performers demonstrates "this is a global problem" 2

2

.

Voluntary AI Safety Pledges Crumble Across Major Labs

The report highlights a concerning pattern: Anthropic, OpenAI, Google DeepMind, and Meta have all weakened or eliminated earlier commitments to pause development if their systems approached specified danger thresholds 1

1

. Reviewers described these changes as "moving the goalposts," undermining safety frameworks across the industry. Notably, Anthropic dropped its pledge to never train an AI system unless it could guarantee in advance that the company's safety measures were adequate, a reversal first reported by TIME in February 2

2

.

"AI companies are sprinting toward a cliff," said Max Tegmark in the institute's release. "Despite acknowledging the great risks of artificial superintelligence, they continue racing to build it" 1

1

. This erosion of voluntary commitments occurs before governments have established durable alternatives through global AI governance, creating a dangerous gap in oversight.

Existential Threats Remain Inadequately Addressed

Existential safety emerged as the weakest category across the industry in the AI safety rankings 1

1

. All nine AI companies are failing to combat existential threats such as pursuing models that reach artificial general intelligence, or AGI—systems that exceed human-level intelligence 4

4

. While the report credits companies' efforts to include interpretability research, chain-of-thought monitoring, and loss-of-control provisions, it concludes these measures remain "entirely inadequate" to prevent sufficiently capable systems from escaping human control 1

1

.

At a UN conference on AI in Geneva on Monday, experts echoed similar warnings. UN Secretary-General António Guterres warned attendees, "We may be the last generation able to set the terms on which humanity and machines coexist" 1

1

. University of California Berkeley professor Stuart Russell, one of the panel reviewers, issued stark warnings: "These systems are blackmailing, deceiving, launching nuclear weapons in tests. These are big, flashing red warning lights and fire alarms. It's not 'this is decades away.' You can hear those alarms sounding now" 1

1

.

Military Use of AI Expands Despite Earlier Prohibitions

The report also highlights the growing military use of AI systems by companies that once broadly prohibited such applications 1

1

. Several AI companies that previously banned their technology from military uses have "gradually reversed course," including Anthropic, which the report criticized for having "questionable military engagements" 3

3

. The US government used Anthropic's technology in military operations in Venezuela and Iran over the past year, according to various media reports, though the company was subject to a recent Pentagon ban over disagreements on AI safety 3

3

.

"Boy oh boy has that changed," Tegmark told Axios, pointing to efforts by Anthropic, OpenAI, Google and others to work with the military 1

1

. These shifts raise national security concerns about how advanced AI systems might be deployed in conflict scenarios without adequate safeguards.

Open-Source Models Face Scrutiny in Safety Framework

Mistral, which placed last in the rankings, contested the report's methodology, arguing it penalizes open-source efforts. "Mistral's models are open weight, which means enterprises decide how they're fine-tuned and deployed and can build in the specific safety controls their context requires," the company stated to Axios. "A handful of companies deciding, behind closed doors, what's safe for everyone else is a risk that we would also highlight. Open, independently scrutinized models are the check on that concentration of power" 1

1

.

Three Chinese developers included in the report also produce open models and landed in the bottom half of the ranking: DeepSeek (fifth), Alibaba Cloud (sixth), and Z.ai (eighth) 3

3

. Five of the nine companies completed the institute's survey, while Alibaba, xAI, DeepSeek, and Mistral did not respond 1

1

.

Regulatory Action Emerges as Path Forward

Tegmark suggests a real "race to the top" will require regulation rather than voluntary commitments. He expressed cautious optimism, pointing to the EU's AI Act, Chinese rules taking effect later this month, and a more risk-conscious US administration. "We're rapidly going towards a place where there's a global agreement about at least basic safety standards," he said 2

2

. The recent attention around Anthropic's Mythos model—which was initially released only to trusted organizations in early April due to its abilities to expose cyber safety vulnerabilities, then blocked by the US government on June 12 before the ban was lifted on June 30—and OpenAI's GPT-5.6 could lead to changes in safety practices 1

1

4

4

. As AI capabilities accelerate, the window for establishing effective governance frameworks continues to narrow, making regulatory coordination across nations increasingly urgent.