The day cameras took down Twitter: the Mirai attack on Dyn
In October 2016 a botnet of hijacked cameras and routers attacked Dyn's DNS service. This one was an attack, not an error, and the devices are still out there.

The devices that broke the internet on 21 October 2016 cost about forty dollars each. They were security cameras, digital video recorders, home routers and network-connected media players, sitting in houses and small offices and shop ceilings, doing exactly what their owners had bought them to do. Their owners had no idea they were participating in anything.
What made them useful to an attacker was not sophistication. It was that each one listened on telnet with a factory default username and password that could not be changed, or that nobody had changed, and that each one had an internet-routable address.
By the end of that Friday, Twitter, Reddit, Spotify, GitHub, Netflix, Airbnb, PayPal and SoundCloud had all been unreachable for substantial periods across North America and Europe, in several distinct waves. None of them had been attacked. The target was Dyn, the company that answered DNS queries on their behalf.
Attack, not error, and the distinction matters here
Almost every large internet outage of the past decade has turned out to be a configuration change, a software defect or an operational mistake. This one was not. Dyn confirmed a distributed denial of service attack and named the Mirai botnet as a primary source of the traffic.
That changes the entire response. An outage caused by error ends when the change is reverted. An attack ends when the attacker stops, or when enough traffic can be filtered, and it can resume. Dyn's day came in waves precisely because there was somebody on the other end deciding when to push again. Merit Network's analysis from its own vantage point describes three distinct phases, with the second, running from roughly 15:55 to 20:40 UTC, carrying global impact.
It also changes who you talk to. Dyn said at the time that it was collaborating in an ongoing criminal investigation and would not speculate about motive or identity.
How DNS turns an attack on one company into an outage for hundreds
DNS is the system that turns a name people type into an address machines can route to. Authoritative DNS, the kind Dyn provided, is the final answer for a given domain. Recursive resolvers at internet providers and public services ask the authoritative server, then cache the answer for a period set by the domain owner, called the time to live.
That cache is why the outage did not arrive everywhere at once. When Dyn stopped answering, resolvers kept serving cached records until those records expired. Sites went dark gradually, in an order determined by TTL values and by which resolver each user happened to be using. From the outside it looked like an intermittent, geographically patchy failure, which is part of why early reporting was so confused.
Then came the second flood. When an authoritative server stops answering, every resolver in the world retries, and retries again. Dyn said the event generated ten to twenty times normal DNS traffic once those legitimate retries were counted. Part of the load crushing Dyn's infrastructure was the internet behaving correctly.
Dyn described the malicious component as masked TCP and UDP traffic over port 53. Mirai also implements a pseudo-random subdomain attack, sometimes called DNS water torture, which asks for names that do not exist. A nonexistent name cannot be served from cache, so every one of those queries has to be forwarded to the authoritative server. It is a way of making a modest volume of traffic expensive to answer, and Dyn confirmed it was used.
Mirai itself
The malware was not clever. That is the interesting part.
Mirai scanned for devices with telnet exposed, tried a table of roughly sixty default credential pairs harvested from device manuals, and loaded itself onto whatever answered. It then killed competing malware on the device, reported in to a command server, and waited for instructions. It lived in memory, so a reboot cleaned a device, and it typically reinfected within minutes because the underlying exposure was unchanged.
Academic analysis presented at USENIX Security in 2017 traced the botnet's growth and composition in detail, and CISA had already published an alert about the heightened threat from Mirai and similar botnets before the Dyn attack, prompted by the attack on the KrebsOnSecurity website in September 2016 that exceeded 620 gigabits per second.
The event that governs attribution came in late September 2016, when Mirai's source code was published publicly. Within weeks there were many separate botnets built from it, run by unconnected people. That is why the question "who attacked Dyn?" has no public answer even though Mirai's authors are known.
Attribution, stated carefully
In December 2017 the US Department of Justice announced guilty pleas from Paras Jha, Josiah White and Dalton Norman in connection with creating and operating Mirai, alongside a separate click-fraud scheme. Jha also admitted a series of attacks against Rutgers University.
Those cases concern the creation and operation of the botnet. No public charge has ever been brought for the attack on Dyn specifically, and the source code release means the Dyn operator may have had no connection to the authors at all. Anyone who tells you confidently who attacked Dyn is going beyond the public record. Theories at the time ranged from a nation state to someone annoyed about a gaming server, and none was substantiated.
Figures worth separating
Reported by Dyn: roughly 100,000 malicious endpoints, traffic at ten to twenty times normal levels including legitimate retries, masked TCP and UDP over port 53, and confirmation that Mirai was a primary source and that the water torture technique was used.
Not confirmed by Dyn: the 1.2 terabits per second peak that appears in most retellings of this incident. Dyn's published analysis did not confirm a peak volume. The figure has been repeated so often that it reads as established fact, and it is not.
From third parties: the lists of affected websites, assembled by journalists and users, and the various phase timings, which differ between analyses depending on the observer's vantage point. CISA's figure of 620 gigabits per second relates to the earlier attack on KrebsOnSecurity, not to Dyn, and the two are frequently conflated.
What is confirmed and what is not
Confirmed: the date, the multi-wave structure, that this was a DDoS attack, Mirai's involvement as a primary source, the techniques Dyn described, the retry amplification, the general pattern of impact, and the DOJ case against Mirai's authors.
Not established: who conducted the attack, the motive, the precise peak traffic volume, the number of end users affected, and any reliable financial total. Dyn was acquired by Oracle shortly afterwards and its original incident write-up is no longer at its old address, which is one reason so much of the detail now circulates secondhand.
What to take from it
Use two independent DNS providers. This is the single control that separated the organizations that stayed up from the ones that did not. Two providers serving the same zone, with records kept in sync, means losing one is a degradation rather than an outage. It is not expensive and it is not difficult, and ten years on a great many organizations still have one.
Know your TTLs before you need them. Short TTLs let you move traffic quickly in an emergency and increase your dependence on your DNS provider being reachable. Long TTLs do the opposite. Pick deliberately, and know what the values currently are, because the default you set years ago is probably still there.
Treat DNS as critical infrastructure in your own planning. It is usually owned by whoever registered the domain, monitored by nobody, and absent from the asset inventory. It is also the single dependency whose failure makes everything else unreachable regardless of how well it is running. Our guide to DDoS attacks covers the defensive side, and third-party risk covers how to assess the provider.
Change default credentials on anything with a network connection, and segment what you cannot change. The devices in this botnet were not servers. They were cameras and recorders bought by people who were never going to read a manual. In an organization, that category includes building access controllers, printers, cameras, displays, sensors and anything else procured outside IT. Put them on their own network segment, block inbound access from the internet, and keep their firmware current. The basics are in keeping business devices updated and network security.
Rehearse the attack-versus-error question. Dyn's incident and Fastly's incident looked similar from outside and required opposite responses. Decide in advance what evidence you would use to tell them apart and who makes the call. That belongs in your incident response plan.
Device asking you to confirm a password after an outage? Fake router and camera "security update" prompts are a common follow-up. Paste the message or link into CyberWatch AI Scan for a free check.
What has aged worst about this incident is not the technology. It is the assumption, widespread in 2016, that an attack of this kind required resources. It required a published list of default passwords and a scanner. The devices are still being sold, still shipping with credentials nobody changes, and still coming online faster than anybody is patching them. The Dyn attack is less a historical event than a demonstration of a standing condition.
Frequently asked questions
What happened to Dyn on 21 October 2016?
Dyn's managed DNS infrastructure was hit by a large distributed denial of service attack in several distinct waves through the day. Because DNS translates names into addresses, websites that relied on Dyn became progressively unreachable as cached records expired, even though the sites themselves were running normally.
Was this an attack or an error?
An attack, unambiguously. Dyn confirmed it was a distributed denial of service event and named the Mirai botnet as a primary source of the traffic. This is the clearest possible contrast with the configuration and software failures that cause most large internet outages.
What was Mirai?
Malware that scanned the internet for devices exposed on telnet, tried a short list of factory default usernames and passwords, and enrolled anything that answered into a botnet. Its victims were consumer internet-of-things devices: IP cameras, digital video recorders, home routers and network media players.
How big was the attack?
Dyn reported roughly 100,000 malicious endpoints and said the event generated ten to twenty times normal DNS traffic once legitimate retries were included. A peak figure of 1.2 terabits per second has been repeated widely for years; Dyn's own published analysis did not confirm a specific peak volume, so treat that number as unverified.
What techniques were used?
Dyn described masked TCP and UDP traffic over port 53, the port DNS uses. Mirai also includes a pseudo-random subdomain attack, sometimes called DNS water torture, which asks a resolver for names that do not exist so the query cannot be answered from cache and must be passed to the authoritative server. Dyn confirmed that technique was used against it.
Why did legitimate traffic make it worse?
When an authoritative DNS server stops answering, recursive resolvers around the world retry. Millions of ordinary resolvers politely asking again, repeatedly, becomes a second flood on top of the attack. Dyn specifically attributed part of the ten-to-twenty-times traffic multiple to these retries.
Who was behind it?
Nobody has ever been publicly charged for the Dyn attack. Three men pleaded guilty in 2017 to creating and operating Mirai, but the Mirai source code had been published publicly in late September 2016, so by October many unrelated people were running their own botnets with it. Dyn declined to speculate, citing an ongoing criminal investigation.
Why were some sites affected and others not?
Organizations that used Dyn as their only authoritative DNS provider had no fallback once Dyn stopped answering. Those with a second, independent provider serving the same zone kept resolving. That single architectural decision largely determined who had a bad day.
Could this happen again?
Yes. The underlying conditions, enormous numbers of internet-exposed consumer devices with weak or unchangeable default credentials, have improved only partially. Mirai variants are still active and its source code remains public.
What is the main defensive lesson?
For organizations, DNS redundancy across two independent providers, and knowing your record TTLs. For device makers and owners, that factory default credentials on anything reachable from the internet are not a minor configuration oversight but the raw material of botnets.
Sources
- Heightened DDoS Threat Posed by Mirai and Other Botnets (TA16-288A), US Cybersecurity and Infrastructure Security Agency
- Hacked Cameras, DVRs Powered Today's Massive Internet Outage, KrebsOnSecurity
- Mirai IoT Botnet Co-Authors Plead Guilty, KrebsOnSecurity
- Justice Department Announces Charges And Guilty Pleas In Three Computer Crime Cases Involving Significant DDoS Attacks, US Department of Justice
- DDoS On Dyn Used Malicious TCP, UDP Traffic, Dark Reading
- The Dyn Attack: Who Participated? Who Was Impacted?, Carnegie Mellon University Software Engineering Institute
- A view on the IoT-enabled DDoS attacks against Dyn from Merit's vantage point, Merit Network
- Understanding the Mirai Botnet, USENIX Security Symposium 2017
- Understanding and Responding to Distributed Denial-of-Service Attacks, US Cybersecurity and Infrastructure Security Agency
- Denial of Service (DoS) guidance collection, UK National Cyber Security Centre


