The Body Language of a Dog About to Bite: Pre-Conflict Signals on Video

10 May 2026 · Rachel Trafford

The Body Language of a Dog About to Bite: Pre-Conflict Signals on Video

"It came out of nowhere." It is the most common thing owners say after a bite, and it is almost never true. Bites are the end of a sequence, not the start of one. The signals are there, they are just fast, subtle, and easy to miss, especially in real time and especially for someone without the training to read them.

This is where video review changes the conversation. When you can return to the seconds before a conflict and look at them properly, the bite that supposedly came from nowhere almost always turns out to have been preceded by a clear, readable build up. This post is about what to look for in that window, why it matters so much in behavioural terms, and why a lot of AI analysis misses it entirely.

Start with the framework, ABC

Most behaviour work is organised around a simple and powerful structure, ABC. Antecedent, Behaviour, Consequence. The behaviour is the thing everyone notices, the bark, the lunge, the bite. The consequence is what happens next, the person backs away, the other dog retreats, the resource is kept. But the antecedent, what came immediately before, what set the whole thing up, is where the real understanding lives.

You cannot change a behaviour by looking only at the behaviour. If a dog bites and all you study is the bite, you have learned almost nothing useful. What changes a case is understanding the antecedent, the trigger, the context, the build up, the conditions that made the behaviour the dog's best available option in that moment. Get the antecedent right and you can often prevent the behaviour entirely, by changing what leads up to it. Miss the antecedent and you are left managing bites after they happen, which helps nobody and least of all the dog.

This is why the seconds before a conflict are not a side detail. In ABC terms they are the antecedent, and the antecedent is the part that actually tells you how to help.

The ladder of escalation

Most dogs work up to a bite through a recognisable sequence, sometimes called the ladder of aggression. The exact framing varies between practitioners, but the principle is consistent. Dogs generally try the low cost, low commitment signals first, and escalate only if those are ignored or do not work.

The early rungs are subtle and easy to dismiss. Looking away and head turns, an attempt to defuse. Lip licking and nose licking out of context. Yawning when not tired. Blinking and half moon eye, the white of the eye showing as the dog tracks a threat without turning its head.

Then comes the freeze, a brief, total stillness that is one of the most important and most missed signals of all.

The middle and upper rungs are more obvious and escalate quickly. Body stiffening, weight shifts, the body going hard and still. A hard stare, fixed and unblinking. Lip curl and showing teeth. Growl and snarl. The snap, a deliberate miss and a final warning. Then the bite.

The clinically crucial point is that dogs whose early signals have not worked, because those signals were punished, ignored, or trained out, may skip rungs and escalate fast. This is where genuinely "no warning" bites come from. Not an absence of warning across the dog's history, but a dog who has learned that the early warnings are pointless and goes straight to the high rungs. Understanding that is itself an antecedent insight. The history is part of the setup.

The antecedent signals worth pausing on

When reviewing the window immediately before a conflict, the five to ten seconds before it ignites, these are the signals worth slowing the footage right down for.

The freeze. A sudden, total stillness, one of the most reliable and most missed precursors. It often lasts a fraction of a second. In real time it reads as nothing, frozen in a single frame it is unmistakable. A dog that freezes is a dog at a decision point.

The hard stare. Fixed, intense focus on the trigger, often with the head lowered slightly. Different from interested attention by its rigidity.

The body stiffen. Watch the weight distribution. A loose, fluid dog and a dog that has gone hard and still are giving completely different information, and the moment of transition between them is the moment that matters most.

The closed mouth. A relaxed dog often has a loose, slightly open mouth. A mouth that snaps shut and tightens is a tension signal.

Weight forward, or weight back. Forward weight over the front end can signal commitment to confrontation, weight back can signal a defensive or fearful posture. Both matter, and they help you tell offensive from defensive aggression, which changes the whole plan.

Tail carriage and movement. Not just up or down, but the quality of the movement. A high, stiff, fast flagging tail is very different from a loose wag, and either can sit alongside arousal.

Why these antecedent signals are so easy to miss, for humans and for AI

There are three reasons these signals get lost, and the third one is the one most people never think about.

They are fast. A freeze can last under a second. A weight shift happens in the gap between one moment and the next. In real time, watching a live interaction, the human eye goes to the big obvious movement and skips the small precursor.

Attention goes to the wrong place. When a conflict is brewing, observers naturally watch the most active dog, or the trigger, or the owner. The precursor signals are often happening in the dog who is about to act, who may be the stiller of the two and the one nobody is watching.

And here is the technical one. A lot of AI video analysis does not actually watch every frame of a video. To save processing, generic systems sample frames, they take a still every so often, perhaps one frame a second or even less, and analyse those snapshots rather than the continuous footage. For most purposes that is fine. For reading pre conflict body language it is a serious problem, because the signals that matter most are exactly the ones that live between the sampled frames. A freeze that lasts half a second, a flick of whale eye, a stiffen that resolves in a moment, can fall entirely into the gap between two sampled stills and simply never be seen. The system is not lying when it later says there was no warning. It genuinely did not look at the frames where the warning was.

This is the difference between generic analysis and analysis built for this work. Our model is designed to treat the pre conflict window as an antecedent that must be examined, not skipped. When a conflict is detected, the analysis goes back to the seconds before and reviews them deliberately and densely, looking specifically for the precursor signals in that window rather than relying on whatever happened to be caught in a sparse sample. And it is held to internal consistency, it cannot report that a conflict came without warning in one place while describing a hard stare and a body stiffen in another. Those two disciplines, refusing to skip the antecedent and refusing to contradict itself about whether warning was present, are what turn a superficial description into a useful assessment.

A note on multi-dog conflicts

Reading the antecedent window is hard enough with one dog and a trigger. In a multi dog conflict, a household scuffle or a group that tips into a fight, it is harder still, because the signals are distributed across several dogs at once and the scene gets busy fast.

The discipline that helps is to refuse the word chaos and instead track each dog individually through the seconds before. Which dog stiffened first, who was staring at whom, which dog was trying to avoid and which was committing. The dog showing avoidance and appeasement is telling you something completely different from the dog who is orienting and committing, and in a multi dog household, knowing which dog was doing which is the whole assessment. Again, this is antecedent work. The fight is the behaviour, but the few seconds before it tell you why it happened and who was driving it.

Where the trainer remains essential

The analysis can surface the antecedent signals, even the fast ones that generic systems skip. What it cannot do is supply the context that gives those signals their meaning. It does not know that this dog has learned to skip its early warnings, that a pain condition is lowering the threshold, that the confined room and the narrow doorway funnelled the dogs together, that this trigger has a history. The antecedent on the screen is only part of the antecedent. The rest lives in the case history the trainer holds. The tool finds the signals, the trainer reads the situation that produced them.

MyCanine360's behaviour analysis treats the pre conflict window as an antecedent to be examined, not skipped, reviewing the seconds before a conflict densely rather than from a sparse frame sample, and tracking each dog individually through a multi dog incident. Every assessment is reviewed and edited by the trainer. [Learn more / try it].

Related reading:

• Multi-Dog Households: Analysing Inter-Dog Dynamics on Video
• Reading Separation Anxiety on Video: What to Look For