Table of Contents Expand

December 10, 2025 – Thirteen major AI companies including Microsoft, Meta, Google, Apple, OpenAI, and Anthropic have received formal letters from a coalition of US state attorneys general warning about the risks of AI "hallucinations" — when models confidently output false or misleading information.

What the Letters Actually Say

The letters highlight three core concerns:

  • AI systems sometimes produce convincingly wrong answers ("hallucinations")
  • These outputs can spread misinformation, affect decisions, or cause harm
  • Companies must implement stronger safeguards and transparency
This is one of the strongest coordinated signals yet that AI safety and accuracy are becoming mainstream regulatory priorities.

The 13 Companies That Received Letters

CompanyMain AI Product(s)
MicrosoftCopilot, Azure AI
MetaLlama, Meta AI
GoogleGemini, Bard
AppleApple Intelligence
OpenAIChatGPT, GPT-4o
AnthropicClaude
AmazonBedrock, Q
NvidiaAI infrastructure
IBMWatsonx
AdobeFirefly
Stability AIStable Diffusion
PerplexityPerplexity AI
xAIGrok

Why AI "Hallucinations" Are a Serious Issue

Real-world examples already exist:

  • Legal bots citing fake court cases
  • Medical AI suggesting incorrect treatments
  • News summarizers inventing facts
  • Customer service bots giving dangerous advice
Even a 1% hallucination rate becomes unacceptable when AI is used billions of times daily.

How Companies Are Responding

Most companies issued similar statements emphasizing:

  • Ongoing work to reduce hallucinations
  • Confidence indicators (showing when AI is uncertain)
  • Source citations and fact-checking features
  • Human-in-the-loop for high-stakes use cases

What Happens Next

This coordinated action suggests:

  • Increased scrutiny on AI accuracy claims
  • Potential future requirements for "confidence scoring"
  • More pressure to show measurable hallucination reduction
  • Stronger differentiation between entertainment and professional AI tools
The era of "it's just AI being creative" as an excuse is ending. Accuracy is becoming a core product requirement.

Data Sources & Methodology (as of Dec 10, 2025):

  • Official letters released by coalition of state attorneys general
  • Company statements from Microsoft, Meta, Google, OpenAI
  • Reuters, The Verge, TechCrunch coverage
  • AI safety research from Anthropic, OpenAI, and academic sources