What experts are saying about AI Safety
AI safety and alignment — the risks, governance, and technical work aimed at making advanced AI reliable and controllable.
“OpenAI's image generation models are designed to refuse prompts requesting the style of a living artist.”
- Sam Altman
“OpenAI does not currently possess a conscious AI model or one capable of self-improvement.”
- Sam Altman
“Signal's position is that the UK's proposed mandatory on-device nudity scanning constitutes a dangerous mass surveillance system that could be weaponized by governments to scan for political speech.”
- Meredith Whittaker
“OpenAI has a "preparedness framework" that outlines its process for testing models for catastrophic risks before release.”
- Sam Altman
“According to its system card, Anthropic has implemented interventions in Fable-5 to limit its effectiveness on tasks related to frontier LLM development, such as building pre-training pipelines or ML accelerator design.”
- The host of AI Daily Brief