A Guidelight AI Standards study reported by Reuters assessed how major AI companies prepare for systems that may become difficult to control. It covered OpenAI, Anthropic, Google, Meta and xAI.

OpenAI and Anthropic received the highest marks, but the study still found weaknesses across the industry. Areas of concern included independent oversight, monitoring and plans for containing advanced systems.

Safety grades are one piece of evidence, not a complete picture of every product. For customers deploying agents today, access controls, human approval and reliable audit logs remain the most immediate protections.

What this means

  • The study found containment gaps across leading AI labs.
  • OpenAI and Anthropic received the strongest grades in the report.
  • Customers still need their own operational safeguards.

Read the announcement

We explained the update in plain English. Use the original source below for the full announcement and any later changes.

Read the original source