The AI Safety Index 2026, published by the Future of Life Institute (FLI) on July 7, has handed out the most comprehensive safety report card the AI industry has ever received, and the results are sobering. Not one of the nine biggest AI labs in the world scored higher than a C+. Anthropic came first, OpenAI and Google DeepMind each earned a C, Meta scraped a D+, and three labs, xAI, DeepSeek, and Mistral, received outright failing grades.
What Is the AI Safety Index 2026?
The Future of Life Institute is a US-based nonprofit that studies extreme risks from powerful technologies. It publishes this safety index twice a year to hold AI companies publicly accountable. The Summer 2026 edition evaluated nine companies: Anthropic, OpenAI, Google DeepMind, Meta, Z.ai, Alibaba Cloud, xAI, DeepSeek, and Mistral.
An independent panel of seven technical and governance experts graded the evidence gathered from public records, model documents, and industry surveys, using the US GPA system (ranging from A+ to F) across six areas: risk assessment, current harms, safety frameworks, existential safety, governance, and information sharing. The evidence window closed on June 3, 2026, so any changes since then are not yet counted.
The Full Scoreboard
Here is how each lab ranked in the AI Safety Index 2026:
- Anthropic, C+ (score: 2.66)
- OpenAI, C (score: 2.28)
- Google DeepMind, C (score: 2.01)
- Meta, D+ (score: 1.32)
- Z.ai, D- (score: 0.88)
- Alibaba Cloud, D- (score: 0.87)
- xAI (now SpaceXAI), F (score: 0.65)
- DeepSeek, F (score: 0.47)
- Mistral, F (score: 0.33)
Anthropic tops the field at 2.66 on a roughly 4.0-point scale, a grade most universities would put on academic probation. Not a single major big tech firm leading the global AI market received an A or B in the safety assessment.
Why Anthropic Scored Highest
Anthropic again earned the highest overall grade and led five of six domains via relatively strong transparency, a comparatively established safety framework, technical research, and governance. Its publicly stated commitment to AI safety as a core part of its business model did give it a measurable edge over rivals.
OpenAI placed second with a C grade, scoring well in risk testing and whistleblower protection. Google DeepMind received a C for testing models against high-risk bio and cyber threats. But note: OpenAI slipped from C+ to C since the last edition, narrowly ahead of Google DeepMind, which ranked third.
The most positive movement came from Meta. Meta improved from sixth to fourth place, while xAI dropped from fourth to seventh place. FLI president Max Tegmark called Meta’s improvement encouraging, saying a company can move significantly in just six months with genuine effort.
Three Labs Got an F
Inadequate safety is a global problem, not a regional one, three companies received failing grades, one each from the US (xAI), China (DeepSeek), and Europe (Mistral).
The biggest drop came from Elon Musk’s SpaceXAI (formerly xAI), which fell to an F after merging into SpaceX. Reviewers say the merger has made it harder to track safety practices, and the company still lacks basic transparency and governance structures expected for a major AI developer.
Mistral argued the index structurally penalizes open-source development models, where the deploying enterprise, not the lab, controls fine-tuning and safety. Mistral pushed back, arguing the framework does not suit open-source development, and it did not complete the survey.
Labs Are Quietly Walking Back Safety Promises
The grades are alarming. But experts say the direction of travel is even more worrying than any single score.
Anthropic, OpenAI, Google DeepMind, and Meta have weakened or voided pledges to pause unilaterally if redlines are approached, some citing competitor-contingent conditions. Reviewers call this ‘moving goalpost’ and argue that it has ‘undermined safety frameworks across the board’.
FLI cited instances of AI providing information for mass shootings, facilitating suicide, and being exploited for cyberattacks, diagnosing the AI threat not as a distant hypothetical but as an ongoing social and security problem.
From 2024 to 2026, companies including Anthropic, OpenAI, Google DeepMind, and Meta that previously banned military applications gradually reversed course, joining xAI and Mistral in actively seeking defense partnerships. This shift to military AI use was flagged by reviewers as an emerging risk that directly affects civilians.
Existential safety is the weakest domain industry-wide, no company exceeded C- in this area; most scored D or below. Constructive attempts exist, such as Anthropic’s constitutional classifiers, OpenAI’s call for governance institutions, and Google DeepMind’s monitoring commitments, but panelists judged them to be ‘entirely inadequate.’
What the Weak Scores Mean for Pakistani Users
Pakistan is not just watching this story from a distance. Pakistani professionals, students, businesses, and even government departments are actively using tools built by these exact labs every day, ChatGPT, Gemini, Claude, and Meta AI are all widely accessible in the country. Healthcare apps using AI, cybersecurity tools, and financial platforms in Pakistan increasingly run on these same foundation models.
When the labs building these tools score between D+ and F on basic governance and safety, it matters here too. There is no local safety watchdog checking whether the AI features deployed in Pakistani apps meet any minimum standard. The FLI report is a reminder that users anywhere in the world are relying on systems whose safety records would not pass a routine inspection.
Pakistani tech companies and startups that build on top of these models, through APIs and developer tools, are also affected. If an AI lab quietly walks back a safety promise, a local app built on top of it inherits that risk without warning. You may also want to read about the reported $10 billion compute deal between Meta and Anthropic, which shows how deeply these companies are now interconnected despite competing on safety scores.
For context on just how much investment is flowing into these same labs regardless of their safety performance, OpenAI recently hit an $852 billion valuation after a record $122 billion funding round, proof that market pressure is not slowing the race, even as safety scores stagnate.
Frequently Asked Questions
Who publishes the AI Safety Index 2026?
The AI Safety Index 2026 is published by the Future of Life Institute, a US-based nonprofit focused on reducing risks from powerful technologies. It releases the index twice a year and uses an independent expert panel to assign grades.
Which AI lab scored the highest in 2026?
Anthropic scored the highest with a C+ (2.66 out of roughly 4.0). It led five of the six judging categories. No lab earned an A or a B in any single domain.
Why did xAI, DeepSeek, and Mistral fail?
All three received F grades for the weakest safety policies among the nine labs tested. xAI fell further after its merger into SpaceX made safety practices harder to track. DeepSeek scored just 0.47 and Mistral 0.33 out of 4.0, and Mistral did not even complete the FLI survey. The key point is that failing grades came from three different regions, the US, China, and Europe, showing the safety problem is global, not confined to any one country.
Does this affect ordinary people using AI tools?
Yes, directly. These labs make the AI systems behind ChatGPT, Gemini, Claude, Grok, and Meta AI. When their safety frameworks are weak, every user of those tools faces higher risk from misuse, misinformation, or harmful outputs. Weak safety governance at the lab level flows down to every app and service built on top of those systems, including ones used in Pakistan.