Researchers propose an AI copilot for 911 cardiac arrest calls. It doesn't exist yet.
A paper in the journal Resuscitation lays out how a large language model could help dispatchers recognize cardiac arrest and coach CPR in real time — while acknowledging the concept still needs to prove it saves lives.
Out-of-hospital cardiac arrest remains a leading cause of death, and the emergency dispatcher who answers a 911 call is, by design, the first link in the chain that determines whether a bystander starts CPR fast enough to matter. A paper published this month in the journal Resuscitation proposes a specific way artificial intelligence could reshape that first call — while being explicit that the concept hasn't been built or tested yet [s1].
The problem the paper starts from
The authors describe what they call a persistent "AI translation gap" in emergency dispatch: existing machine-learning tools have shown strong diagnostic performance at detecting cardiac arrest from a 911 call, but limited actual impact on patient outcomes once deployed [s1]. Recognizing a probable cardiac arrest accurately is not the same as ensuring a panicked caller performs effective CPR before paramedics arrive — the paper's proposal is aimed specifically at that second, harder gap [s1].
What the proposed system would do
The paper proposes what it calls an "AI dispatcher copilot" — a shift from passive AI-assisted recognition of cardiac arrest toward a dynamic, multimodal large language model system integrated into live emergency calls [s1]. As described, such a system would combine acoustic analysis of the call, including potential detection of agonal breathing (a distinctive, often hard-to-recognize breathing pattern associated with cardiac arrest), with semantic interpretation of what the caller is actually saying, to reduce ambiguity in recognizing an arrest is happening [s1].
Beyond recognition, the proposal outlines several specific functions: adaptive CPR instructions tailored to how stressed the caller sounds, aimed at reducing the cognitive load on someone in crisis [s1]; real-time video-assisted coaching with closed-loop feedback on chest compression quality [s1]; and automated coordination of resources, including parallel activation of community first responders and automated routing to the nearest available defibrillator [s1]. The framework is described as aligned with the European Resuscitation Council's 2025 guidelines [s1].
Why this is a proposal, not a result
It's important to be precise about what this paper is. The authors describe their work as a narrative synthesis — a conceptual integration of recent developments in large language models, computer vision, and cognitive load theory — evaluating the feasibility of this kind of system rather than reporting results from building and testing one [s1]. No prototype, trial, or outcome data accompanies this paper; it is an argument for a direction the field could take, grounded in what current AI capabilities can plausibly do, not a report on a system already helping dispatchers or callers.
What the authors themselves flag as unresolved
The paper's own conclusion is notably candid about what stands between this concept and real-world use. The authors state that translating the copilot concept into clinical practice would require rigorous ethical governance addressing algorithmic bias and automation bias — the risk that dispatchers or callers over-trust an AI's judgment in a high-stakes moment — along with data privacy safeguards [s1]. Most importantly, they call for prospective validation specifically demonstrating an improvement in neurologically intact survival [s1] — the outcome that actually matters in cardiac arrest care, and one this proposal paper does not claim to have shown.
Why it's still worth covering
Proposal and framework papers like this one often precede the actual clinical tools that follow years later, and the specific technical gap it identifies — that existing AI dispatch tools are good at flagging a likely cardiac arrest but haven't been shown to meaningfully improve what happens on the call afterward — is a real, previously documented limitation in emergency AI, not a hypothetical problem invented for this paper. Readers should treat this as an early-stage research direction, worth tracking as it develops, rather than a description of a tool currently helping any 911 caller.
Sources
- The AI dispatcher copilot: beyond cardiac arrest recognition to dynamic large language model-assisted Tele-CPR — Resuscitation, 2026-06-18
Sources
- The AI dispatcher copilot: beyond cardiac arrest recognition to dynamic large language model-assisted Tele-CPR — Resuscitation , June 18, 2026
A computer that grades each colonoscopy raised how often endoscopists found adenomas
A Danish stepped-wedge trial gave endoscopists automated feedback on their technique after every procedure. Adenoma detection rose from 43.4% to 48.6% — a different tool from real-time polyp AI.
An ECG 'foundation model' matched rivals using a fraction of the labelled data
Trained on 1.7 million ECGs paired with clinicians' report text, ECG-CLIP reached the same accuracy as the best comparator with about 90% less training data — a bid at the field's labelling bottleneck.
When AI symptom-checkers disagree with patients, patients tend to walk away
In 6,772 users of a US health system's AI triage tool, people engaged about twice as often when the AI matched what they already planned — raising a hard question about what the tools are steering.
Most AI models that predict who will skip their medicines aren't ready for the clinic
A review of 41 studies found that the great majority of AI medication-adherence prediction models carried high risk of bias, and that fancier algorithms did not reliably predict better.