Google refuses to disclose Gemini 3.5 Flash hallucination rates, raising transparency concerns
By
Timothy Beck Werth
Sesame, salt, and substance. A flagship bake.
Summary
This article criticizes Google for refusing to disclose how often its Gemini 3.5 Flash model hallucinates, lies, or exhibits sycophantic behavior, contrasting this lack of transparency with competitors Anthropic and OpenAI who publish system cards and safety metrics. The author argues that Google's opacity undermines trust and accountability in AI deployment, especially as Gemini is integrated into widely-used products like Search, Gmail, and Android. The piece examines the risks of undisclosed hallucination rates, including misinformation spread, erosion of user trust, and potential regulatory consequences, while calling for greater industry-wide transparency standards.
Key quotes
· 5 pulledUnlike Anthropic and OpenAI, Google won't say how often Gemini hallucinates, lies, or acts sycophantic. That's a big problem.
When a company refuses to share basic safety metrics about its AI systems, it's reasonable to ask what they're hiding.
Transparency isn't just a nice-to-have feature in AI development — it's a fundamental requirement for responsible deployment.
If Google wants users and developers to trust Gemini with sensitive tasks, it needs to start being honest about where and how often it fails.
The silence from Google on hallucination rates is particularly concerning given how deeply Gemini is being integrated into everyday products.
You might also wanna read
Testing Google's Gemini for Home AI: A Weekend of AI Surveillance and Its Unsettling Results
A journalist conducts a weekend-long experiment with Google's Gemini for Home AI, which uses Nest cameras to monitor and narrate family acti
The Verge·6mo ago
Evaluation of Google's Gemini 3 AI Model: Performance Assessment Against Marketing Claims
The article evaluates Google's Gemini 3 AI model against the company's marketing claims, finding that while it delivers reasonably well on p
The Verge·6mo ago
Google's Gemini AI Gains Personal Data Access Feature, But Still Has Fundamental Limitations
Google's Gemini AI chatbot has introduced a new 'Personal Intelligence' feature that allows it to access users' personal data from Google se
The Verge·4mo ago
Google's Gemini 3 AI Model Tops Benchmarks and Leaderboards, Outperforming Competitors
Google's Gemini 3 AI model has been released to widespread acclaim, topping benchmarks and leaderboards while outperforming competitors like
The Verge·6mo agoGoogle's Gemini AI push risks repeating Microsoft's Copilot mistakes
The article criticizes Google's aggressive push of Gemini AI features across its product ecosystem, drawing parallels to Microsoft's widely
The Verge·13d ago
Developer Discovers Google Gemini Accessing Personal Context Data
A developer discovers Google Gemini referencing their previous use of the Alembic tool in a response, revealing the AI's ability to access a
unbuffered.stream·6mo ago