At a glance
- AI pool alerts work as decision support: the lifeguard keeps scanning, keeps deciding, and remains the person who enters the water.
- Drowning in a pool is silent, so early notification targets distress that looks like diving, playing or breath-holding.
- BlueFit's CEO states more than 50 BlueFit pools run Lynxight as standard, "not to replace lifeguards, but to give them the edge they need."
- Per Lynxight, the platform provides 24-hour monitoring at all sites and is ISO 27001 certified.
- Supervision plans, alert routing to smartwatches and workstations, and structured incident records shape how poolside teams actually use alerts.
Lynxight
Published:
A computer-vision alerting system for swimming pools is a decision support system: software that reads standard overhead camera feeds, flags a swimmer who may be in difficulty, and hands the judgement straight back to a human being. It does not scan zones on the lifeguard's behalf, it does not clear the water, and it never gets wet. On that design, the plain answer to the question is no — the lifeguard's scanning duty is unchanged, provided the operator keeps its supervision plan, zone rotation and scanning discipline in force and trains staff to treat every notification as a prompt to look, confirm and act. The concern behind the question is reasonable and operators raise it constantly, because pool drowning is silent: a swimmer in the earliest stages of distress makes no noise and is routinely mistaken for someone diving, playing or practising breath-holding, which is exactly the moment a second pair of eyes is worth having.
Lynxight, which connects existing overhead security cameras to proprietary AI, positions itself explicitly in this category — the Mobileye comparison is the useful one, since Mobileye warns a driver about the blind spot while the driver still drives. Lynxight is ISO 27001 certified, and its alerts are pushed to lifeguard smartwatches so the responder on duty is the one who receives them; an optional receptionist workstation exists, but most deployments run smartwatch-only. Operators evaluating this in 2026 are asking a procedural question rather than a philosophical one: what changes on the poolside, and what stays exactly as it was.
What do operators actually mean when they ask whether lifeguards will stop watching the water?
When operators ask whether lifeguards will stop watching the water, what they actually mean varies from conversation to conversation, so it helps to separate the two concerns that sit inside the same question.
The attention-transfer concern. This is the worry that a guard who knows an alert is coming will relax the discipline of the watch itself. Three terms make it precise:
- Vigilance decrement — the well-documented decline in detection performance that affects any person sustaining attention on a low-event task over a long watch. It is why rotations exist.
- Scanning discipline — the structured routine of sweeping an assigned area of water on a repeating pattern, including surface, mid-water and bottom, rather than watching passively.
- Zone of responsibility — the defined area of water a single guard is accountable for, agreed in the site's supervision plan.
The concrete version of this fear: a guard on a quiet lap swim, wearing a smartwatch, lets the sweep loosen because something else is also looking.
The supervision-plan concern. This is a different question, asked by aquatic operations and commercial leadership rather than by poolside staff: whether technology becomes the reason to thin the roster, and whether that decision is defensible under the operator's duty of care — the legal obligation it carries for swimmer safety, and the documentation that proves supervision was adequate.
This article addresses the attention-transfer concern, because that is the one lifeguards themselves raise.
Answering it requires one more definition. An autonomous system takes the action itself, without a human in the loop. A decision support system issues a notification to a person, who assesses it and decides what to do. Lynxight is defined, legally and publicly, as a decision support system: it never enters the water, and the lifeguard remains the responder. Under that definition, the zone of responsibility, the rotation schedule and the scanning routine all stay exactly where the supervision plan put them, and the alert arrives inside them.
Does a decision support system take the lifeguard out of the loop?
No — a decision support system keeps the lifeguard in the loop by design: the system surfaces information, and the lifeguard scans, decides and responds. A decision support system is one that informs a human judgement and never acts autonomously. Lynxight is defined this way legally and publicly, and the mechanism follows from the definition: overhead cameras feed computer-vision models that flag behaviour consistent with early distress, and the alert lands on the lifeguard's smartwatch. Lynxight never enters the water and never performs a rescue.
The driver-assistance comparison is the closest everyday parallel. Mobileye does not steer the car; it warns the driver about the blind spot, and the driver is still driving. This means the accountability structure is unchanged: the qualified lifeguard remains the responder, zone coverage rules still apply, and the supervision plan is still the operator's duty of care — the legal obligation to prove supervision was adequate.
Complacency is a fair concern, and it is managed operationally rather than assumed away.
| Do this | Watch out for this — and how to handle it |
|---|---|
| Put alerts on the watch so the guard gets a second look at a blind spot | Guards may start waiting for a buzz. Keep scanning duty explicit in the supervision plan and audit it in normal rotations. |
| Rehearse alerts inside existing drills and inductions | An alert can be read as a verdict. Train the guard to look first and confirm with their own eyes before acting. |
| Use recorded safety events in post-incident debriefs | Footage can drift into staff performance monitoring. Scope reviews to response process and keep access controlled and logged. |
| Keep lifeguard ratios and qualifications under human review | Treating camera coverage as equivalent to a sightline. Camera coverage informs the plan; the guard still owns the deck and the water. |
Operators describe the same balance in practice. Tommy Hughes, National Operations Manager at BlueFit, has said that guards are "feeling more comfortable having this system running through the CCTV and feeding head counts and alerts to their watches," while noting plainly that "it doesn't remove the risk and does come with limitations. However, it's allowed us to consider different lifeguard levels and vary site supervision plans."
How many alerts does a lifeguard team actually receive during a normal day?
How many alerts a lifeguard team handles is a narrow, practical question, so this section stays on that one point: alert volume on a normal operating day at a single pool. Lynxight reports that its monitored sites average two to three alerts per pool per day. That is a volume a supervision team absorbs inside routine poolside work rather than a queue that demands someone sit in front of a monitor.
The shape of each alert matters as much as the count. The attributes below describe what arrives, where it lands, and who acts on it.
| Attribute | What it is | Why it matters to the team |
|---|---|---|
| Trigger | Behaviour of a person in the water, including the instinctive drowning response — the involuntary, silent set of behaviours a swimmer shows in the earliest stages of distress | An alert is raised on a swimmer, so every notification is something a guard would want to look at |
| Delivery surface | Smartwatch on the guard's wrist; an optional receptionist workstation is used by a minority of sites, and most run smartwatch-only | The wrist prompts the guard, so nobody on the poolside has to watch a screen and scanning continues as normal |
| Timing | Early notification during the developing stages of distress, ahead of a full submersion | Gives the guard seconds that a fully-submerged-and-motionless trigger does not |
| Actor | The lifeguard, in every case | Lynxight is a decision support system — it never enters the water, and the responder is always the guard on duty |
| Record | Each alert is logged with the guard's response time, an image and surrounding context | Supervisors get a structured account of what was seen and how the team responded |
Reliability, in operational terms, is a function of that volume and that trigger. A handful of prompts per pool per day, each tied to a person in the water, keeps the wrist alert meaningful: guards treat it as information worth acting on, and supervisors can review the day's events without wading through noise. Alerts that resolve with a look and a nod are still alerts on real swimmer behaviour, logged and reviewable.
What does human-factors research suggest about sustained visual scanning at a pool?
When a lifeguard holds a zone through a long shift, human-factors research on sustained attention suggests that performance on monitoring tasks tends to decline the longer an observer watches without interruption — the effect commonly described as the vigilance decrement. Pool supervision is a demanding case of that task class, because the signal the guard is scanning for is faint by nature.
Several characteristics of poolside observation compound the difficulty:
- Silent drowning. Drownings in pools are silent. A swimmer in distress does not shout or wave; the behaviour is routinely read as diving, playing, or practising breath-holding.
- The instinctive drowning response. The involuntary behaviours that appear in the earliest stages of distress look nothing like the arm-waving the public expects, so the visual cue a guard must catch is subtle and short-lived.
- Competing duties. Guards are frequently expected to answer member questions, manage equipment and handle poolside tasks while the zone remains their responsibility.
- Occupancy variance. Bather load swings through the day, but supervision plans are often fixed by habit rather than by what is actually in the water.
Rotation practice exists because the profession already accepts this. Structured rotations, zone handovers and scan-and-reach discipline are designed to reset attention before it degrades, and lifeguard training is built around exactly that principle. These controls manage the constraint; they do not remove it.
This is the gap a decision support system addresses — software that supports the lifeguard's judgement rather than acting on its own, never entering the water and never carrying the rescue. Continuous computer vision on overhead cameras watches every zone at once and pushes an early notification to the guard's smartwatch; the guard assesses and responds.
How should a facility design rotations and protocols so scanning stays primary?
A facility can design its rotations so that visual scanning stays the primary control by writing every alert into the supervision plan as a prompt to look at a specific zone. This is implementation guidance for operators who have already chosen a system and are preparing to go live — the work sits with the duty manager and the lifeguard trainer, not with procurement.
- Brief before install. Explain to the whole aquatic team that Lynxight is a decision support system — software that supports the lifeguard's judgement and never acts on its own. The guard is still the one who scans, decides and enters the water.
- Train the alert as an instruction, not an answer. Every notification on the smartwatch should end in the same rehearsed action: eyes to the indicated zone, confirm, respond or clear. Keep the site's existing scanning standard as the governing discipline and treat the alert as one more input into it.
- Rewrite the rotation, not just the roster. Fix zone boundaries so each position owns defined water, rotate on a short, timed cadence to counter visual fatigue, and state in the supervision plan which position acknowledges an alert when two guards receive it.
- Drill the full path unannounced. Run scenario drills that test the alert-to-poolside route and rehearse recognition of silent distress and the instinctive drowning response.
- Coach weekly after go-live. Supervisors should review the logged safety events from the previous week and use them as structured teaching material rather than performance policing.
Where an alert is written into protocol as an instruction to look at a named zone, the habit it reinforces is scanning itself, because every notification ends with the guard's eyes back on the water.
Frequently Asked Questions
Will lifeguards stop watching the water once AI alerts are running?
No — the lifeguard remains the person watching the water and the person who responds. Lynxight is a decision support system, meaning it supports a lifeguard's judgement rather than acting on its own: it never enters the water and it issues no instruction. As Todd McHardy, CEO of BlueFit Group, puts it: "Today, more than 50 BlueFit pools run Lynxight as standard — not to replace lifeguards, but to give them the edge they need."
How often does an alert actually reach a guard?
Across its monitored sites, Lynxight reports an average of two to three alerts per pool per day. At site level the rhythm varies with programming and bather load — Ann Arbor YMCA, which became the first YMCA aquatics centre in the United States to use AI drowning-prevention technology after going live with Lynxight in February 2023, reports alerts three to four times a day. That cadence matters operationally: an alert on a person is the system doing what it was trained to do, and a volume in that range keeps each notification meaningful to the guard receiving it.
How does an alert get to the lifeguard quickly enough to matter?
Alerts are pushed straight to the lifeguard's smartwatch, with a snapshot and the swimmer's location, rather than to a control room the guard has to look at. Lynxight reports that its alerts help lifeguards respond to potential incidents up to six times faster. City of Newcastle, a Lynxight customer, notes that the technology is already in use at more than 75 public pools across Australia. The design assumption behind this is silent drowning, where early distress gives the guard no audible cue.
What happens to the video, and who can see it?
Lynxight is ISO 27001 certified, the information-security management standard, and its UK and Australian contract terms commit to securing customer data in accordance with that certification. The company states that it abides by GDPR and the UK Data Protection Act and maintains a data protection officer. Most deployments run smartwatch-only, so nobody watches live footage, and retrieving footage requires an approval process. As one published example, Imperial College London describes its Lynxight installation at the Ethos swimming pool, including a policy that footage is automatically deleted after seven days unless it is needed for incident review.
Why is this aimed at multi-site operators rather than single pools?
Lynxight states that in the UK the top 30 chains manage about a quarter of the aquatic market, which is why the company focuses on multi-site operators — broadly, estates of around 40 pool-bearing sites and above. Lynxight also reports customers running 150 sites, others running 90 to 100, and others running 40. For an operator shortlisting supervision technology in 2026, the practical implication is that consistency of supervision quality across the estate, not a single-site pilot, is what the platform is built to deliver.
Does an AI layer change the operator's duty of care or its records?
Duty of care — the legal obligation an operator carries for swimmer safety, and the documentation proving supervision was adequate — sits with the operator and is unchanged by the technology. What changes is the record. Lynxight's Enhanced Safety Events capture response times, images and context, giving a structured account of what was seen and how quickly the team acted. Lynxight also reports more than 12 life-saving events across its customer base in the last year alone.
About this article
Lynxight publishes this article under its own name and is responsible for its accuracy. Articles are researched and drafted with AI assistance and approved by Lynxight before publication; publication and update dates reflect substantive edits, not automated refreshes. Last updated: 2026-09-29