7 Frontier AI Companies Rated Across 33 Safety Indicators in Future of Life Institute's 2025 AI Safety Index
What happened
The Future of Life Institute published the 2025 AI Safety Index - Summer 2025 in May 2025, evaluating seven leading frontier AI companies against 33 measurable indicators organized across six domains: risk ownership, accountability, independent oversight, safety culture, responsible deployment, and transparency. The index names specific companies and assigns granular performance assessments at the indicator level, making it one of the most detailed third-party benchmarking exercises applied to frontier AI developers to date. Among its concrete findings, the report identifies coordination weaknesses at Google DeepMind, calls out insufficient disclosure practices around third-party model evaluations, and flags the absence of published whistleblowing policies as a systemic gap across multiple assessed organizations. The assessment covers companies operating across multiple jurisdictions and is framed as a global evaluation rather than one tied to any single regulatory regime. The index methodology draws on domains that overlap with requirements emerging from the EU AI Act, the G7 Hiroshima AI Code of Conduct, and the Bletchley Declaration on AI Safety.
Why it matters
- ·Regulatory exposure: The index domains overlap directly with disclosure, auditing, and oversight obligations emerging under the EU AI Act and UK AI governance frameworks, meaning poor index performance by a model provider may signal regulatory non-compliance risk for enterprises relying on those providers.
- ·Operational impact: The explicit identification of gaps in third-party evaluation transparency and independent oversight at named providers creates new due diligence requirements for procurement and vendor assessment processes, particularly for organizations deploying high-risk or general-purpose AI systems.
- ·Organizational risk: The absence of published whistleblowing policies across multiple assessed firms is increasingly treated as a baseline governance deficiency by regulators in multiple jurisdictions, and non-response or failure to remediate index findings by providers could become a material factor in enterprise risk disclosures or board-level AI governance reporting.
Governance controls affected
What to do now
- ☐Map the six domains of the 2025 AI Safety Index against your existing supplier assessment criteria and identify gaps in how you evaluate safety culture and independent oversight maturity for frontier model providers.
- ☐Review contracts and vendor agreements with any of the seven assessed companies to determine whether whistleblowing policy publication and third-party evaluation disclosure are currently required as contractual obligations.
- ☐Incorporate the index findings into your third-party AI risk assessment process, flagging coordination deficiencies and transparency gaps identified for specific providers as elevated risk factors.
- ☐Track whether assessed providers issue public responses or remediation commitments in response to the index findings, and document provider responsiveness as part of ongoing vendor monitoring records.
- ☐Assess whether your organization's EU AI Act compliance documentation for high-risk or general-purpose AI systems adequately addresses third-party provider obligations related to independent oversight and evaluation transparency as highlighted by the index.
What to watch next
Compliance teams should monitor whether the seven assessed companies issue formal responses or updated transparency disclosures in reaction to the 2025 AI Safety Index, as a pattern of non-response may become relevant to vendor risk classifications and regulatory assessments. Regulatory developments under the EU AI Act, particularly the finalization of codes of practice for general-purpose AI models with systemic risk designations, are likely to formalize obligations that mirror several index domains, including independent auditing and incident reporting. Teams should also watch for similar benchmarking initiatives from other civil society organizations and for signals from the UK AI Safety Institute or US AI Safety Institute indicating whether third-party index outputs may be referenced in future supervisory guidance.