Martha Kalifatidis’ viral parenting clip
Content note: This article discusses suicide, online harassment and mental health. Support details are included at the end.
When a crowd starts surging at a concert, responsible organisers stop the show before somebody gets seriously hurt.
Social platforms can see when thousands of people suddenly descend on one person. Yet they let the pile-on continue at full speed.
After the death of Professor Jason Arday, that gap is impossible to ignore.
On 3 August, Professor Simon Baron-Cohen, director of Cambridge University’s Autism Research Centre, posted on X about his friend. He described what was happening to Arday as relentless bullying of a vulnerable autistic man, raised safeguarding concerns and asked for it to stop.
Elon Musk replied with one word: “Liar.”
Eleven days later, Arday was found dead at his home in Battersea. He was 41. Police are treating his death as unexpected but not suspicious, and an inquest will determine how he died. That process is still underway, so we need to be careful about the conclusions we draw.
But we can still look honestly at what happened online. A leading autism researcher publicly warned that a specific person was at risk. He identified the mechanism and asked people to stop. The owner of the platform then amplified the pile-on against him.
Nothing in that chain was designed to slow things down.
I spend much of my working life with creators, but I have also been the talent on the other side of this.
Years ago, I said on Jordanian radio that I disliked a Jordanian national dish. That was it. An opinion about food. The backlash was so intense that I locked myself in a room for three days.
I remember how horrific it felt. I remember the shame and the way I started questioning my own worth because a large group of people had suddenly decided I was awful. When you already carry trauma, a public pile-on does not land as a neat disagreement that disappears when you close the app. Your body can read it as danger.
Now I see versions of that experience happen to creators. I see the sleeplessness, the compulsive refreshing and the way six hours of public contempt can completely rearrange someone’s sense of who they are.
So I want to argue for something more useful than another round of thoughts and prayers about online cruelty.
Social platforms should build a circuit breaker for outrage.
What I Am Proposing
Platforms should offer every user an optional protective mode.
If someone switches it on, the platform would monitor sudden increases in hostile comments against what is normal for that account. The speed of the increase matters more than the raw number.
When hostility rises far beyond that account’s usual pattern, the platform would step in automatically. Notifications and access to the comment inbox could be paused for a short, defined period. A neutral notice could appear in the comments explaining that unusually high activity has triggered a temporary pause.
The account owner would keep control and could override the pause with one click. It would simply give them a temporary layer of protection while the surge is happening.
My first instinct was to make this protection specifically available to people who identify as neurodivergent or as living with serious mental illness. But that creates a new problem: platforms would be storing highly sensitive health information.
The safer version is to offer protective mode to everyone, while allowing people who know they are at higher risk to choose a more sensitive setting. The platform does not need to know why.
Accountability Still Needs Safeguarding
The allegations made against Arday were serious, and some appear to have had substance. Institutions disputed parts of his academic history, and Cambridge opened an investigation into his appointment. He resigned on 5 August, saying the personal cost of the scrutiny had become too great.
At the same time, a Liverpool John Moores University panel had previously found that similarities in his thesis were the result of honest and reasonable error rather than deliberate plagiarism.
A professor’s academic record should be scrutinised. That scrutiny can still be delivered with safeguards.
Mental Health Europe reported that Arday was the subject of 249 articles across 15 UK national outlets in 22 days. That is more than 11 articles a day before counting the social posts, clips and comments generated by each one.
Lord Woolley, principal of Homerton College, described the coverage as “a bloodsport” and said Arday felt hounded. Baron-Cohen said he watched Arday’s mental health deteriorate over the final ten days of his life.
When someone becomes the focus of sustained national criticism, who is checking whether that person is safe?
Right now, the answer is usually nobody.
Some People Carry A Higher Risk
Baron-Cohen raised autism because it can be a risk factor during an experience like this.
The research supports that concern. Autistic adults experience much higher rates of anxiety, depression, self-harm and suicidality than the wider population. ADHD has also been associated with a significantly higher risk of suicide attempts and death.
Many neurodivergent people use the term “rejection sensitive dysphoria” to describe the intense pain of perceived criticism or rejection. It is not a recognised clinical diagnosis, so the science around the term remains unsettled. We do know that many people with ADHD experience difficulties with emotional regulation, and that social exclusion can register as genuine emotional pain.
Suicide is complex and rarely has a single cause. The research does show a foreseeable risk, especially for groups already experiencing higher rates of anxiety, depression and suicidality. That should be enough for platforms to take prevention seriously.
For Creators, This is Part of Their Job
Arday was an academic. For creators, online hostility is ambient weather.
In one study of 135 creators across Instagram, TikTok and YouTube, almost every participant remembered experiencing hate or harassment. For one in three, it was a regular occurrence. Some said it made them censor themselves or leave platforms entirely.
The risk rises for people who are already marginalised. A 2024 study of disabled creators found that 96% had experienced disability-related hate or harassment and 78% had experienced sexual harassment. LGBTQ disabled creators faced an even greater risk of ableist abuse.
Pew Research found that 41% of US adults had experienced online harassment, including around seven in ten lesbian, gay or bisexual adults.
These are the same people platforms celebrate in diversity campaigns while leaving them to manage the personal cost of visibility alone.
The Comment Section Runs 24 Hours A Day
An Australian coronial finding into the death of teenager Jessica Cleland described how mobile access exposed her to upsetting material around the clock and allowed her to reread it.
Rereading is important.
The harm of a pile-on is rarely one comment. It is unlimited access to an accelerating stream of contempt at 2am, on a device you cannot put down, with a red badge counting every new hit.
Research involving people under 25 has found a strong association between cybervictimisation, suicidal thoughts and suicide attempts. It gives us another reason to build a brake into the system.
Content Moderation Only Goes So Far
Content moderation only addresses part of the harm.
In August 2025, a 57-year-old obstetrician known publicly as Dr Shao died in Zhoukou, China. Her family said she had endured cyberbullying for more than seven months.
According to Douyin, 15 accounts had posted 89 videos about her. The platform removed or restricted 76 of them, deleted 457 comments and approved 32 of the 48 reports submitted by Shao and her representatives.
On paper, the moderation system was working. Reports were filed. Content was removed. Rules were enforced.
But every part of that system focused on the content. Almost none of it focused on the state of the person receiving it.
The same pattern appears repeatedly: support arrives after the damage. Suicide has many contributing factors, and coroners have rightly resisted simplistic conclusions about the role of online abuse. Platforms can respect that complexity while still acting on a foreseeable risk.
We Already Stop Events When Crowds Become Unsafe
At concerts and sporting events, organisers are expected to intervene when crowd movement becomes dangerous. They might stop the performance, turn up the lights or ask people to step back.
The pause responds to the speed and direction of the collective movement. It does not need to judge every person in the crowd.
An online pile-on works in a similar way. One comment may not be dangerous. Thousands of comments arriving within minutes can become something entirely different.
The same three principles should apply online:
The trigger should be clear and consistent.
Activity should be measured against what is normal for that account.
The pause should be temporary and designed to interrupt a dangerous surge while criticism can continue elsewhere.
We already accept that responsibility in physical spaces. It should not disappear when the crowd moves online.
What Would Trigger The Pause?
There is no magic percentage at which criticism becomes a pile-on. Any threshold would need to be tested rather than presented as settled science.
Research into online firestorms does show that sudden surges can be detected quickly. One study identified a major event in less than 80 seconds by looking at five-minute windows of negative and total comment activity. Other research has detected firestorms before human observers recognised that they had begun.
A platform could compare the current rate of hostile comments with the account’s usual activity at the same time and day. It would act only when the increase sat far outside that normal range and there were enough comments to show it was a genuine surge.
A practical starting model could look like this:
Alert: Hostile comments at least double, with a meaningful minimum number of comments across two five-minute windows. The user receives a prompt offering protective mode.
Protective pause: Hostile comments at least triple and remain unusually high. Notifications and the comment inbox pause for 30 minutes, with human review and a one-click override.
Severe surge: Hostile comments rise at least fivefold, or continue climbing rapidly. The pause extends to two hours and triggers a rapid safety review.
These numbers are a starting point for testing. A platform would need to trial them privately before launch and publish how often the system gets it wrong.
Minimum numbers matter. An increase from one hostile comment to two is technically a 100% rise and means almost nothing. A jump from 10 to 30 hostile comments in a few minutes is different. Without a minimum volume, the feature would constantly misfire on smaller accounts and quickly become useless.
Credible threats, doxxing and self-harm content should bypass these thresholds entirely and go straight to human review.
The Notice Must Protect Privacy
The public notice is where my own thinking changed.
My first instinct was to tell commenters that the person had entered protective mode. But that could turn a private health disclosure into a public label, and a public label can become another target. It would also create a record of vulnerability inside an advertising business.
The notice should describe the activity while leaving the person’s private information out of it:
Comment activity here is unusually high, so comments are temporarily limited. This is not a finding of wrongdoing.
That slows the social momentum without disclosing anything about the account holder.
Three rules should be non-negotiable. Protective mode must never influence advertising or recommendation systems. It must never be visible to other users. And it must be easy to switch off, with any related data deleted.
The Objections Are Real But Solvable
Automated systems are imperfect. They struggle to distinguish hostility from sarcasm, slang, heated debate or culturally specific language. Research has also found that some hate-speech systems are more likely to flag African American English as offensive, while toxicity scores do not translate consistently across languages.
Built carelessly, a circuit breaker could misread a creator’s community and restrict the very person it was designed to protect.
That risk needs proper testing across languages, dialects, genres and account sizes. Platforms should publish false-pause rates, provide human appeal and give creators control over every trigger.
Right now, the main intervention is deleting a comment after the person has already read it 40 times.
Some people will see this as paternalistic. Making the pause opt-in, temporary and easily overridden keeps control with the person receiving the comments.
Platforms Have Already Built Most of The Parts
Platforms have already built most of the technology required.
Instagram’s Limits can hide comments and message requests during spikes of unwanted attention. Twitch’s Shield Mode lets broadcasters activate several chat protections at once. X has tested Safety Mode, Reddit has Crowd Control, and YouTube can hold comments for review. Instagram also reminds users to take a break.
The missing piece is the automatic trigger.
At the moment, the switch sits on the inside of the door, and the person expected to reach it is the one being buried.
There is also a policy path. Australia’s eSafety Commissioner already asks platforms to anticipate and design out foreseeable harm through its Safety by Design framework. An outrage circuit breaker is exactly the kind of preventative feature that approach should encourage.
Inclusion has to Mean Designing for Who is Actually in the Room
The creator economy talks constantly about authenticity, vulnerability and bringing your whole self online. Then it hands people a system with no brakes and files the damage under “personal resilience”.
When we tell a neurodivergent creator to leave public life if they cannot take the abuse, we remove them from the industry and then call their absence a lack of talent.
In Arday’s case, somebody warned a platform in public that a vulnerable person was being buried by it. The platform’s owner called him a liar.
When a physical crowd becomes unsafe, somebody stops the show. Online, people should get the same protection.
If this article brought anything up for you, support is available in Australia through Lifeline on 13 11 14 and Beyond Blue on 1300 22 4636.
Sources and Reading
The Jason Arday case
Cambridge autism expert claims bullying of Jason Arday was relentless, The Observer — Baron-Cohen's 3 August post, the Musk reply, and his account of Arday's final ten days
Tributes paid to Jason Arday, The Guardian — the Battersea death, his age, and the Met's "unexpected but not suspicious" wording
Jason Arday: Cambridge professor in plagiarism row quits, BBC News — the 5 August resignation and the personal cost quote
The rise and fall of Jason Arday, The Guardian — the disputed academic history and the Liverpool John Moores panel finding
When scrutiny exposes a double standard, Mental Health Europe — 249 articles across 15 outlets in 22 days
Calls for public inquiry into Jason Arday "witch hunt", BBC News — Lord Woolley's "bloodsport" and "hounded" comments
Higher risk groups
Anxiety and depression in adults with autism spectrum disorder, Psychological Medicine
Risk markers for suicidality in autistic adults, JAMA Network Open
Suicidal behaviour among persons with ADHD, British Journal of Psychiatry
Rejection sensitive dysphoria, Cleveland Clinic — that it is not a recognised clinical diagnosis
Emotion regulation in adults with ADHD, systematic review, PLOS ONE
Neural correlates of social exclusion, meta-analysis, Rotgé et al.
Creators and harassment
How creators experience and cope with hate and harassment online, Google Research — the 135 creator study
Ableism experienced by disabled content creators, CHI 2024 — the 96% and 78% figures
The 24 hour comment section
Cyber bullying claimed our Jessica, ABC — the coronial finding on round-the-clock access and rereading
Self-harm, suicidal behaviours and cyberbullying in young people, John et al.
Dr Shao and the limits of moderation
Douyin vows to crack down on cyberbullying, China Daily — the 15 accounts, 89 videos, 457 comments and 48 reports
The death of Doctor Shao puts online platforms' obligations to the test, Global Times — the family's account of seven months of abuse
Detecting a surge
Understanding online firestorms, Pfeffer, Zorbach and Carley — the five-minute windows and sub-80-second detection
Against Others: Understanding Online Firestorms, Zhang et al. — detection before human observers
Where automated detection fails
The Risk of Racial Bias in Hate Speech Detection, Sap et al.
The Biases of Toxicity Detection Models, Röttger et al. — that toxicity scores do not translate across languages
Existing tools and policy
Protecting our community from abuse on Instagram, Meta — Limits and spikes of unwanted attention
Wellbeing and safety tools, Meta — the break reminders
