I was in Croatia when the alert went out. The first I knew about it was a flurry of messages from friends back home, then the memes, then a few clients asking whether they should be doing anything differently. Nobody’s phone in our group had gone off, because none of us were on a UK network at the time, and there had been no wider information campaign to catch us. If you happened to be abroad, or between networks, or simply asleep, the entire strategy for reaching you was a single push notification. That gap has stuck with me more than the alert itself, and it’s a big part of why I wanted to write about this properly rather than just add to the pile of hot takes.
For anyone who missed it: on the evening of Friday 14 August, the government sent an Emergency Alert to phones across England and Wales, warning of a very high risk of wildfires and asking people not to use disposable barbecues, naked flames or fireworks. It ran on a rolling basis from around 7pm until 11pm. It came after a genuinely serious few days, with homes destroyed in Stourbridge and firefighters tackling dozens of wildfires across the country. Whatever else is true about this alert, the underlying risk was real.
What followed was a live and fairly even split of opinion, visible everywhere from LinkedIn to the comment threads under the trade press coverage. Some people thought this was exactly what the system exists for. Others felt it stretched the “genuine emergency” threshold the Emergency Alerts system, and the law behind it, is meant to be reserved for. I don’t want to land on one side of that. I think the more useful thing, for people who might one day have to make a similar call in their own organisations, is to open up the parts of this that don’t get much airtime.
Was it Proportionate?
The legal mechanics are worth setting out plainly, because a lot of the online debate skips past them. The Emergency Alerts system runs on an exemption written into the Privacy and Electronic Communications Regulations, which were specifically amended to allow public authorities to ask mobile networks to broadcast an alert where an emergency, as defined in the Civil Contingencies Act 2004, has happened, is happening, or is about to happen. Wildfire risk is explicitly one of the stated use cases. So the “was this legal” question has a fairly clean answer: yes. The more interesting question, and the one actually being argued about, is whether a general preventative warning sent to millions of people, most of whom were never near a fire, sits comfortably within the spirit of a system built for immediate, localised threats to life.
There’s a reasonable case on both sides. In favour: the conditions were as serious as the UK has seen in years, the request came directly from Fire and Rescue Services rather than being a political decision made in isolation, and a single careless disposable barbecue really can start a fire that costs someone their home. Against: this wasn’t tied to one incident or one evacuation, it went out nationally rather than to the areas actually at risk, and several people who’d lived through the alert told the trade press it felt more like a public information campaign than an emergency warning. Both of those things can be true at once, and I don’t think the answer is obvious either way.
A single channel doing all the work
This is where the Croatia story becomes relevant to more than just me. As far as I can tell, there was no sustained public information effort around this alert, no advance notice, no adverts, nothing beyond a press release posted shortly before 7pm and the alert itself. For anyone reachable by a UK 4G or 5G signal at the right moment, that’s fine. For anyone who wasn’t, including a fair number of second-home owners, tourists, people between contracts, or simply anyone whose phone was off, the alert may as well not have existed.
That’s not really a criticism of the alert being sent. It’s a question about what sits behind it. A genuinely resilient warning system doesn’t rely on one channel doing the entire job. It has fallbacks, and it assumes a meaningful proportion of the intended audience won’t get the primary message. I don’t know what the right answer looks like at national scale, but I know what it looks like at the scale most of us actually work at: if your organisation’s crisis communication plan has exactly one route to your people, you don’t have a plan, you have a hope.
The bilingual delivery problem
My parents were in Cheshire when the alert went out. They received it twice: first in Welsh, then some time later in English. Neither of them speaks Welsh. In the gap between the two messages, they genuinely didn’t know whether to be worried or not, which is a strange thing to sit with when your phone has just made the noise that’s specifically designed to make you worried.
The reported gap between the Welsh and English broadcasts was around four minutes, based on the timings recorded on the government’s own Emergency Alerts website, which is shorter than what my parents experienced. I’d rather flag that discrepancy honestly than pretend the four-minute figure and their experience are the same thing. It may be a network or handset variation, or it may simply be that four minutes feels a lot longer when you don’t know what the message means.
What struck me most is that this isn’t a new problem. The same publication’s reporting notes that the very first national test of this system, back in 2023, ran into the same issue, with an English message going out ahead of a bilingual Welsh version, and evidence later showing the system couldn’t reliably handle two simultaneous messages. A subsequent test in September 2025 tried a different approach, sending an English-only message with a link to a Welsh translation, which the Welsh Language Commissioner found breached the Cabinet Office’s own bilingual communication requirements. New technology intended to fix this was reportedly due to arrive with contracts starting in April 2026. Friday’s four-minute gap raises a fair question about whether that fix is actually working yet, or whether we’re still one design decision away from getting this right.
For anyone building multi-language alerting into their own crisis plans, the lesson isn’t “translate faster.” It’s that sequencing itself is a design decision with consequences. A gap of any length between two versions of the same warning creates a window where part of your audience is anxious, confused, or simply left guessing, and that window doesn’t feel small to the people standing in it.
The Safeguarding Gap
This is the part of the story I think deserves more weight than it’s had. Women’s Aid, amongst others, has highlighted that the alert put some domestic abuse victims at risk. Many survivors rely on a second, hidden phone to stay in contact with support services or to plan an escape, and an unannounced siren-style alert is exactly the kind of thing that can give a hidden device away to an abuser in the room. Women’s Aid’s head of external affairs put it simply: survivors “frequently turn to second phones to help them access lifesaving support or flee from abuse,” and the charity is now asking government to give advance warning that alerts might be issued, wherever that’s possible.
What makes this harder to wave away is the timing. By most accounts, domestic abuse charities had almost no advance notice, somewhere in the region of fifteen minutes, and no real opportunity to get safeguarding guidance out to the people who needed it before the siren went off. A charity director working in this space described the core problem well: safety systems that don’t consult the organisations supporting the most vulnerable people risk becoming a threat to exactly those people.
I don’t think this makes the alert wrong. I think it makes the process around the alert incomplete. Building safeguarding into a blunt, one-way broadcast tool is genuinely difficult, because the entire value of the tool is speed and reach, and speed is precisely what safeguarding consultation costs you. But “difficult” isn’t the same as “not worth doing,” and fifteen minutes’ notice to the charities best placed to advise on this doesn’t look like a system that’s tried very hard yet.
Alert fatigue and the cost of crying wolf
A number of people reported receiving the alert more than once, including overnight, and several said afterwards that they’d switched emergency alerts off on their phones as a result. That’s the part of this story that should worry anyone who works in resilience, because it’s not really about this alert at all. It’s about the next one.
One of the system’s own architects was quoted saying the decision had “really divided opinion,” and that government needs to be clearer about the criteria for when the system will and won’t be used. I think that’s right, and I’d go further: every time a warning system is used in a way a meaningful chunk of the public sees as disproportionate, it erodes trust that took years to build and can be gone quickly. If enough people have quietly turned notifications off by the time something genuinely immediate happens, the system has failed at the one moment it exists for, regardless of how well it performed on paper this time.
The thread underneath all of this
None of these four issues, proportionality, single-channel reliance, bilingual sequencing, or safeguarding, are unique to a national government system with a legal exemption behind it. They’re the same questions any organisation should be asking about its own mass-notification and crisis communication arrangements, just at a smaller scale and with lower stakes.
Do we have more than one route to reach people, and have we actually tested the fallback rather than assumed it works? If we communicate in more than one language, have we tested what happens when the translation takes longer than expected, rather than assuming it will arrive alongside the original? Have we thought about who might be put at risk by the communication itself, not just by the thing we’re warning them about, and have we built in even a few minutes’ consultation with the people who’d know? And do we have a clear, written sense of what does and doesn’t meet our own threshold for using the loudest tool in the box, so that we’re not making that call from scratch, under pressure, every single time?
I don’t think there’s a tidy verdict to reach on Friday’s alert. The risk was real, the intent was good, and several of the things that went wrong were things the system has apparently gone wrong on before. That combination, a genuine emergency met with a response that repeats known weaknesses, is a more useful thing to sit with than picking a side. If you’re the person who’ll have to make this kind of call for your own organisation one day, I suspect it already sounds familiar.
