







The voice safety classifier supports 30 languages and checks for profanity, asking for PII, harassment, sexual content, dating and romantic, discriminatory speech, illegal and regulated content, and disruptive audio. github.com/roostorg/model-community/tree… You can test it out here! huggingface.co/spaces/Roblox/voice-safety-cl…
model-community/roblox-voice-safety-classifier at main · roostorg/model-community
github.comAug 19, 2026 at 5:14 PM
Move Over, Love Languages: Do You Know Your 'Route of Safety'?
According to this expert, these "trauma-informed love languages" can lead to deeper connections.

Google's call-scanning AI could dial up censorship by default, privacy experts warn | TechCrunch
A feature Google demoed at its I/O confab yesterday, using its generative AI technology to scan voice calls in real time for conversational patterns
47 U.S. Code § 230 - Protection for private blocking and screening of offensive material
No provider or user of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider.

voiceome.org
Find information, resources and relevant links for voiceome.org. This domain may be for sale.
Are LLMs Stifling Political Speech? An Assessment of How AI Models Protect Free Expression | Oversight Board
The Oversight Board’s first evaluation of large language models (LLMs) shows that some of the world’s most-used models from Anthropic, DeepSeek, Google, Meta
Uberi/speech_recognition
Speech recognition module for Python, supporting several engines and APIs, online and offline.
Refusal in Language Models Is Mediated by a Single Direction
Conversational large language models are fine-tuned for both instruction-following and safety, resulting in models that obey benign requests but refuse harmful ones. While this refusal behavior is widespread across chat models, its underlying mechanisms remain poorly understood. In this work, we show that refusal is mediated by a one-dimensional subspace, across 13 popular open-source chat models up to 72B parameters in size. Specifically, for each model, we find a single direction such that erasing this direction from the model's residual stream activations prevents it from refusing harmful instructions, while adding this direction elicits refusal on even harmless instructions. Leveraging this insight, we propose a novel white-box jailbreak method that surgically disables refusal with minimal effect on other capabilities. Finally, we mechanistically analyze how adversarial suffixes suppress propagation of the refusal-mediating direction. Our findings underscore the brittleness of current safety fine-tuning methods. More broadly, our work showcases how an understanding of model internals can be leveraged to develop practical methods for controlling model behavior.

A reporting checklist for large language models in behavioural science
Large language models offer new opportunities for behavioural science, but their rapid evolution poses challenges for research rigour. We introduce a consensus-based reporting checklist to improve transparency, reproducibility and ethical accountability of large-language-model-based research in the behavioural sciences.

Jailbreaking Large Language Models: If You Torture the Model Long Enough, It Will Confess!
A Cautionary Tale…

The Consent Layer: Using ligatures to make web text expensive to scrape without asking
ShieldFont is an open-source creative technology project that offers a practical opt-out from unauthorized AI training and disrupts what is collected when that choice is ignored. It swaps 45.8% of content words (around 24.4% of all words) in a page's source code for other (partially) random words, while the font restores the original text on screen. Readers see the work as intended; mass scrapers collect an altered version. In testing, shielding caused over 90% of pages that would otherwise pass the quality filter to be rejected, keeping them out of the training pipeline. Of those that still passed, 19.4% of all words conveyed false meaning, adding noise to unauthorized AI training datasets. This paper's goal is to walk newcomers through the whole process, in plain language and in order: the project's rationale, how it was built, the results, how to deploy it, and where to contribute.
Underdog Woof
Our text-to-speech models: one voice in 31.7 MB, or two voices in 40.6 MB with fewer word errors than Fish Audio Drama 3 at over 1,000 times smaller. They read phone numbers, dates, prices and codes the way people say them.

BotPsy: On-Device Personal Assistant with a Sub-Billion-Parameter Agentic Language Model
A Blog post by QVAC on Hugging Face

PII-TRACE: Detecting Personal Data Before It Leaves a Device
Discover how PII-TRACE tests recurring PII detection in long multilingual conversations with a compact 0.6B local model.

Local AI, together
Moss now turns speech into text on your own computer, and Presence uses it so a whole call gets transcribed with nobody’s voice leaving their machine. Any Moss tool can use it.

When I worked in software dev, I was taught to check three non-English languages when testing text strings: German to test the largest possible version of a string, Japanese or Chinese to test the shortest possible version, and Thai to check the TALLEST possible version.
The forced global adoption of puritanistic American cultural norms vis sexuality and nudity via app store/payment processor rules is one of the least-discussed free speech issues of the last decade and this lack of attention is going to hurt online expression so much in the years to come.
Bee 🐝
I think also we cannot ignore the potential social and cultural ramifications of normalizing the demonification of anything adult or sexual The whole system is stacked against any sort of movement in the other direction and it's not something the porn industry can handle alone at this point