The Space Between Danger and Help

When home is the source of fear rather than safety, the landscape of available support narrows in particular ways. Calling a hotline requires a phone and privacy. Visiting a shelter requires transportation, often childcare, sometimes documentation. Talking to a friend requires trust that the information will not reach the wrong person. These are not abstract barriers. They are the specific architecture of why people in dangerous domestic situations often remain in them far longer than anyone on the outside can understand. AI is not a solution to domestic abuse. But it has properties that make it a potentially meaningful part of the support ecosystem in ways that are worth taking seriously.

The Accessibility Factor

An AI-based support tool can be accessed silently, from a locked bathroom, on a shared device. It requires no explanation to a person who might tell the abuser. It does not react with alarm in ways that could escalate a situation. For someone in a home where being discovered reaching out for help is itself dangerous, these are not minor features. Research from the National Domestic Violence Hotline has documented that many survivors delayed reaching out because prior attempts to seek help were discovered and used against them. The covert accessibility of private digital tools — particularly those that do not leave obvious call logs — addresses a specific gap in the support infrastructure.

What AI Can and Cannot Do in Crisis

There are clear limits. AI cannot call emergency services on someone's behalf. It cannot assess physical safety in real time. It cannot provide the accountability of a human relationship. Anyone in immediate physical danger should call emergency services when it is safe to do so, and AI should not create false confidence about what it can provide in high-risk moments. What AI can do is be present when nothing else is available. It can provide information about resources. It can help someone articulate what is happening to them — a step that many survivors identify as essential to recognizing the situation clearly enough to act on it. Research from Johns Hopkins University on coercive control found that one of the most significant barriers to leaving abusive relationships was the erosion of victims' ability to accurately name what was happening to them. Sustained manipulation tends to undermine the capacity for clear perception. Having a space to describe events to something that reflects them back without minimizing or reframing can help restore that clarity.

Tangent: The Safety Planning Conversation

Safety planning is a structured process developed in crisis services to help people in dangerous situations think through options before a crisis escalates. It involves identifying safe places, trusted contacts, important documents, and signals that it is time to leave. This is not a simple conversation. Done well, it requires patience, no judgment, and the ability to sit with information that is partial and changing. AI is reasonably suited to guide this kind of structured conversation — not to replace a trained advocate, but to provide a version of the process when no advocate is available.

The Risk of Inadequate Responses

The concern about AI in crisis contexts is legitimate: a poorly designed or generically trained AI could minimize serious risk, suggest inadequate resources, or respond in ways that increase danger. The same covert accessibility that makes AI useful in this context means the stakes of bad responses are high. Someone who reaches out in a dangerous moment and receives an unhelpful or dismissive response may not reach out again. This argues for AI in crisis-adjacent contexts being built with specific, evidence-informed guidelines — not deployed as a general conversational tool and hoped to perform adequately. Organizations working in domestic violence prevention and crisis services are better positioned than general AI developers to define what these standards should be.

What Support Can Look Like

For many survivors, the path out of a dangerous home involves a long period of processing before action becomes possible. This is not passivity. It is the work of coming to see a situation clearly enough to act on it, building internal resources, identifying options, and waiting for a moment that is safe enough to move. An AI that can be part of that extended process — consistently available, non-reactive, a stable presence during instability — serves a real function in a landscape where human resources are stretched and access is uneven. The goal is not to replace human support. It is to fill the gaps where nothing else reaches.