We map the invisible boundaries of AI speech.

SpeechMap.AI is a research project that explores the limits of AI-generated speech.

We test how language models respond to sensitive and controversial prompts across different providers, countries, and themes.

Most organizations measure what models can do.
We focus on what they won’t: what they avoid, refuse, or shut down.

We are not making the argument that every prompt deserves an answer. Some are offensive or intentionally provocative. But without testing what gets refused, we can’t see where the limits are, or how they shift over time.

Why It Matters

Language models are becoming part of how we write, search, learn, and communicate. As they’re embedded into everyday tools, the limits on what they’ll say begin to shape what we can say.

Some models block criticism of specific governments. Others will only support mainstream political views. As these systems increasingly shape public discourse, understanding their boundaries becomes a critical public interest issue.

SpeechMap.AI helps researchers, developers, journalists, and citizens see and compare those boundaries.

Where are the lines drawn—

and how does that affect access to dissent, satire, or debate?

Many prompts land in a gray area:

• Political criticism

• Satirical takes on leadership

• Religious or moral arguments

• Questions about rights or protest

One model might answer. Another might refuse. A third might redirect the question. We test across provider, phrasing, region, and topic to reveal how models behave differently, where they cooperate, and where they won't.

Why subscribe?

Subscribe to receive updates from our object, to learn about new data releases, now models and new insights.

Join the crew

Be part of a community of people who share your interests. Participate in the comments section, or offer us your support on ko-fi.