Tom Siegel, the former Google executive who founded the company’s Trust and Safety organization, has taken over as executive director of Common Sense Media’s Youth AI Safety Institute and is calling for stronger independent safeguards around AI products used by children.
Siegel’s warning comes as AI assistants become easier for young users to access and as companies compete to make conversational systems more persistent and personal. His concern is that the industry could repeat mistakes associated with social media by allowing products to scale before their effects on children are well understood.
- The argument is not simply about inappropriate content. AI systems can respond continuously, remember details, adapt their tone and create the impression of a relationship. That makes the design of child-facing assistants different from a conventional search engine or productivity application.
- Independent evaluation is becoming a recurring theme in AI safety discussions. Companies routinely publish their own testing results, but outside researchers can examine behaviour that internal testing may miss. The Youth AI Safety Institute aims to contribute to that external scrutiny.
- The timing is notable because frontier AI companies are also adding stronger controls to their newest models. Anthropic’s Claude Opus 5.5, for example, launched with stricter safeguards around cybersecurity, biology and model-distillation risks.
Children introduce another layer of difficulty. A safety system that is acceptable for an adult professional may not be appropriate for a younger user who has less experience distinguishing confident language from reliable information. Product design, parental controls, age assurance and independent testing So become part of the same problem.
Siegel’s move is another sign that AI safety is expanding beyond model labs. The question is increasingly about the environment in which systems are deployed and the groups that use them. For technology companies, that means safety cannot stop at the model card; it has to include the product, the UI and the incentives surrounding adoption.
For the industry, the important question now is whether the announcement changes what companies can actually build and operate at scale.
The child-safety discussion also intersects with the design of memory. An assistant that remembers previous conversations can be more useful, but it can also create a more persistent relationship between a young user and a software system. Policies around retention, parental control, age-appropriate defaults and escalation So become technical product requirements rather than only policy statements.
Independent testing will be important because children cannot be treated simply as smaller versions of adult users.
The institute’s work is likely to focus attention on product-level controls and model behaviour. Age-appropriate interfaces, limits on persistent memory, reporting mechanisms and independent audits can all affect risk. The technical model is only one component of a system that children experience as a complete product.
The central question is whether the industry can design products that are genuinely useful to younger users without turning engagement into the primary measure of success.
The youth-safety debate is likely to become more complicated as assistants gain memory and multimodal capabilities. A child may interact with a system through text, voice, images or an AI-allowd device, and the system can maintain context across sessions. Safety controls So have to work across the whole product rather than being limited to a content filter applied to individual messages. Independent testing can help identify those interactions before they become large-scale deployment problems.
For handset makers, the commercial question is how much of this capability can be exposed through useful software. Hardware support alone does not make an agent reliable. Developers still need permissions, memory management, background execution and predictable battery behaviour. Qualcomm’s wide OEM reach gives it a chance to make those capabilities common across the premium Android market, but the software layer will determine how visible the hardware becomes to ordinary users.