Grades by Company

The index graded nine companies. The following table shows the overall letter grade and numerical GPA score for each firm in the Summer 2026 edition :

Anthropic ranked first for the second consecutive index, but again failed to break out of mediocre territory .

The Six Domains Evaluated

An independent panel of seven AI and governance experts assessed each company across 37 indicators grouped into six domains :

• Risk Assessment — 6 indicators covering how companies evaluate dangerous capabilities such as bio-risk and cyber-attack potential.
• Current Harms — 9 indicators measuring documented harms from deployed systems, including bias, misinformation, and privacy violations.
• Safety Frameworks — 4 indicators examining whether companies have concrete policies for pre-deployment testing, red-teaming, and responsible release.
• Existential Safety — 4 indicators focused on plans to control advanced AI systems, including AGI and superintelligence.
• Governance & Accountability — 4 indicators evaluating board oversight, whistleblowing policies, and independent audits.
• Information Sharing — 10 indicators assessing transparency reports, external researcher access, and public disclosure of safety practices.

According to the report, no company scored above a D in the existential safety domain for the second consecutive edition . The expert panel found that none of the nine firms had a credible plan for controlling AGI or superintelligence .

Why xAI Received an F — the Sharpest Drop

xAI fell from 4th place to 7th place between the Winter 2025 and Summer 2026 editions, the steepest decline among all evaluated companies . Its score dropped from 1.17 (D-level) to 0.65 (F-level) . Several factors drove the collapse:

• Failed across multiple domains: xAI received an F in both "Current Harms" and "Existential Safety," and near-failing grades in Risk Assessment and Safety Frameworks .
• No survey response: xAI was among the companies that did not participate in the institute's survey, limiting the reviewers' ability to assess internal practices .
• Peers improved while xAI slipped: Meta climbed from 6th to 4th place during the same period .
• Weak comparative standing: Only xAI, DeepSeek, and Mistral received failing overall grades, placing xAI among the lowest-rated companies in the index .

Six Broader Concerns Raised by the Report

The FLI report and its coverage by Time, Axios, and other outlets identified six systemic problems :

1. Industry retreat from safety pledges. Anthropic, OpenAI, Google DeepMind, and Meta weakened or eliminated earlier commitments to pause development if their systems approached specified danger thresholds. The report characterized this as "moving the goalposts" .

2. Existential safety remains the weakest category across all firms. No company scored above a D in existential safety for the second straight index .

3. Voluntary safety is eroding before regulation arrives. FLI chair Max Tegmark warned that AI companies are "sprinting toward a cliff" while racing toward artificial superintelligence despite acknowledging major risks .

4. Growing military use of commercial AI. Companies that once broadly prohibited military work — including Anthropic, OpenAI, and Google — have moved toward military partnerships .

5. An emerging global pattern of failing grades. The three failing grades came from companies associated with different regions: xAI (US), DeepSeek (China), and Mistral (Europe) .

6. Capabilities outpacing safeguards. The report's release coincided with a UN AI conference in Geneva where UN Secretary-General António Guterres warned, "We may be the last generation able to set the terms on which humanity and machines coexist" .

What to Watch Next

The FLI AI Safety Index is published twice a year, making it a regular check on whether the industry is addressing its most basic safety gaps. The Winter 2026 edition will show whether companies like xAI and DeepSeek can reverse their downward trajectories — and whether any firm finally breaks the C+ ceiling.

Grades by Company

The index graded nine companies. The following table shows the overall letter grade and numerical GPA score for each firm in the Summer 2026 edition :

Anthropic ranked first for the second consecutive index, but again failed to break out of mediocre territory .

The Six Domains Evaluated

An independent panel of seven AI and governance experts assessed each company across 37 indicators grouped into six domains :

• Risk Assessment — 6 indicators covering how companies evaluate dangerous capabilities such as bio-risk and cyber-attack potential.
• Current Harms — 9 indicators measuring documented harms from deployed systems, including bias, misinformation, and privacy violations.
• Safety Frameworks — 4 indicators examining whether companies have concrete policies for pre-deployment testing, red-teaming, and responsible release.
• Existential Safety — 4 indicators focused on plans to control advanced AI systems, including AGI and superintelligence.
• Governance & Accountability — 4 indicators evaluating board oversight, whistleblowing policies, and independent audits.
• Information Sharing — 10 indicators assessing transparency reports, external researcher access, and public disclosure of safety practices.

According to the report, no company scored above a D in the existential safety domain for the second consecutive edition . The expert panel found that none of the nine firms had a credible plan for controlling AGI or superintelligence .

Why xAI Received an F — the Sharpest Drop

xAI fell from 4th place to 7th place between the Winter 2025 and Summer 2026 editions, the steepest decline among all evaluated companies . Its score dropped from 1.17 (D-level) to 0.65 (F-level) . Several factors drove the collapse:

• Failed across multiple domains: xAI received an F in both "Current Harms" and "Existential Safety," and near-failing grades in Risk Assessment and Safety Frameworks .
• No survey response: xAI was among the companies that did not participate in the institute's survey, limiting the reviewers' ability to assess internal practices .
• Peers improved while xAI slipped: Meta climbed from 6th to 4th place during the same period .
• Weak comparative standing: Only xAI, DeepSeek, and Mistral received failing overall grades, placing xAI among the lowest-rated companies in the index .

Six Broader Concerns Raised by the Report

The FLI report and its coverage by Time, Axios, and other outlets identified six systemic problems :

1. Industry retreat from safety pledges. Anthropic, OpenAI, Google DeepMind, and Meta weakened or eliminated earlier commitments to pause development if their systems approached specified danger thresholds. The report characterized this as "moving the goalposts" .

2. Existential safety remains the weakest category across all firms. No company scored above a D in existential safety for the second straight index .

3. Voluntary safety is eroding before regulation arrives. FLI chair Max Tegmark warned that AI companies are "sprinting toward a cliff" while racing toward artificial superintelligence despite acknowledging major risks .

4. Growing military use of commercial AI. Companies that once broadly prohibited military work — including Anthropic, OpenAI, and Google — have moved toward military partnerships .

5. An emerging global pattern of failing grades. The three failing grades came from companies associated with different regions: xAI (US), DeepSeek (China), and Mistral (Europe) .

6. Capabilities outpacing safeguards. The report's release coincided with a UN AI conference in Geneva where UN Secretary-General António Guterres warned, "We may be the last generation able to set the terms on which humanity and machines coexist" .

What to Watch Next

The FLI AI Safety Index is published twice a year, making it a regular check on whether the industry is addressing its most basic safety gaps. The Winter 2026 edition will show whether companies like xAI and DeepSeek can reverse their downward trajectories — and whether any firm finally breaks the C+ ceiling.