Across two iPhone launches this fall, Cequence stopped the largest malicious bot surge of the season at some of the world’s largest mobile carriers. That surge hit days before a pre-order opened rather than during it. Carriers kept their shop, cart, checkout, and activation APIs running through both events, and October brought the first launch season where AI shopping agents showed up alongside malicious bots.
Key takeaways
- Go-live demand arrives as a spike rather than a bigger day.
- The biggest bot surge came before the launch.
- Tens of thousands of IPs traced back to a handful of fingerprints.
- Cequence blocked a small fraction of launch traffic, and real customers got through.
- AI agents such as Meta's Muse now shop alongside people.
Two iPhone launches put carrier bot management to the test
The new iPhone lineup opened for pre-order on September 12 and reached stores on September 18, and a second device, iPhone Duo, opened for pre-order on October 16. Attackers planned around both dates. At one carrier, an abuse campaign against a trial-ordering flow appeared for the first time the day before the September pre-order opened, then returned for the October launch; at another, the biggest attack of the season landed two days ahead of the October pre-order. Fraud operators now plan around launch windows the same way marketing teams do.
Human demand concentrated in the first minutes of go-live
Daily totals barely moved on launch days. At one carrier, full launch days landed between 42 million and 65 million requests, inside the normal range.
The demand showed up in the minutes after pre-orders opened, when request rates doubled within three to four minutes, from about 520 to more than 1,000 requests per second. Across launch-morning windows, total traffic rose a moderate 12% to 29% while peak request rates rose 73% to 119%.
Demand also moved between APIs as each launch progressed, with each phase driving traffic to a different part of the business:
| Launch phase | What customers did | Biggest surge seen |
|---|---|---|
| Pre-order opening | Browsed catalog, pricing, and offers; filled carts | Catalog traffic up 13x from overnight levels in the opening hour; offers up 4x |
| In-store availability | Tracked orders, verified numbers, and activated lines | Phone line provisioning up 16x; number verification up 7x; order lookup up 5.6x |
| Second launch (October) | Opened the app, signed in, consented, and shopped | App launch and sign-in paths up 40% to 49%; cart up 32% |
For capacity planners, sizing infrastructure to daily volume misses the event. Sizing it to the go-live peak, with bot management removing automated traffic from that peak before it reaches origin, is what keeps checkout and activation running.
Credential stuffing and promotion abuse peaked before the October launch
September: real shoppers drove the launch
Legitimate demand dominated the September launch. At one carrier, traffic rose 23% in the launch-morning window while blocked requests (malicious traffic) fell 26%, and the baseline morning a week earlier contained more automation than launch morning did. At another, blocked requests held between 0.11% and 0.18% of traffic across all four launch windows, in line with a normal day.
October: 22 million blocked requests in one day
The October launch drew far more automation. Two days before pre-orders opened, Cequence blocked 22.05 million malicious requests in a single day at one carrier, about 11 times its daily average of 1.99 million. Blocking stayed elevated for three days, then returned to normal on launch day.
| Target | Requests blocked | What the attack looked like |
|---|---|---|
| Loyalty promotion | 10.66M | The same user IDs arrived from 31,900 IPs but only 24 device fingerprints, about 145x the prior week |
| Account sign-in | ~4.5M | Credential stuffing from a handful of toolkits rotating through about 3,000 IPs |
| Session start | 2.36M | Burst traffic tripping per-endpoint rate limits |
| Guest login | 2.05M | Two fingerprints spread across 2,600 IPs |
In each case, a small number of bot toolkits rotated through proxy IP pools to pass as crowds. IP reputation alone would have counted tens of thousands of different visitors; behavioral fingerprinting reduced that to two dozen.
Behavioral fingerprinting exposed attacker camouflage
On October pre-order day, the attackers traded volume for camouflage. A session-abuse campaign that used 6 IPs a week earlier spread across 153,651 IPs on launch day, at roughly two requests per IP, far below the threshold of any IP rate limit. A trial-order abuse campaign spread across 45,599 IPs, about 10 times the prior week, to blend into genuine launch traffic.
Cequence caught both campaigns by matching fingerprints and behavior, while legitimate launch traffic continued normally.
AI shopping agents showed up in the October launch
October was also the first major launch after Meta released Muse, a personal AI agent that signs in to a user's accounts, compares plans, fills carts, and checks out on the user's behalf. OpenAI's always-on Dots agents followed weeks later. Within two weeks of Muse's release, Cequence saw Muse-matching traffic at more than half of the customers it studied, and that traffic was present during the October launch.
These agents complicate launch-day bot management in two ways. Muse presents itself as an ordinary Chrome browser, so an agent buying a phone for its owner and a scalper bot hoarding launch inventory can send near-identical requests. Some agents are also customers, which means blocking every agent turns away real buyers on the busiest sales day of the year, while allowing every agent gives bad actors a new attack channel.
During a launch, the question becomes which agent is acting, for whom, and to what end. Cequence answers it with Agent Trust, which keeps a live inventory of the agents in traffic, including ones that don't identify themselves (such as Meta Muse), and checks agent identity against trusted issuers. Agent Trust also uses behavioral fingerprinting to spot a stolen agent token and applies policy to the individual transaction rather than banning the agent outright.
What bot management saved carriers in infrastructure and staff time
Infrastructure that didn't have to scale for bots
During the three-day October surge, Cequence blocked about 30.8 million requests against a normal expectation of about 6 million, so roughly 25 million bot requests never reached origin. On the peak day, blocks ran 11 times normal while total analyzed traffic moved 3%. The carrier's commerce, identity, and activation services didn't autoscale to serve attackers, and capacity stayed available for the go-live burst of real customers.
A war room that monitored traffic instead of responding to incidents
Carriers still ran a launch war room, but the Cequence CQ Prime team and Technical Success Managers monitored each launch window alongside carrier teams and tuned fingerprint-based policies as attack campaigns shifted. When the pre-launch surge hit, Cequence contained it at the edge and prevented it from becoming an all-hands incident.
Holiday bot attacks will follow the same pattern
The same playbook now moves from phone launches into retail, starting with Black Friday on November 27 and Cyber Monday on November 30 and running through gift-card and returns season into early next year. Expect holiday bot attacks to repeat the launch-season patterns at retail scale:
- Attackers will time campaigns around the event, not just during it, hitting promotions, loyalty programs, and gift cards in the days before a sale opens.
- Doorbuster drops will compress demand into minutes, and bots will try to claim items first.
- Low-and-slow campaigns spread across tens of thousands of IPs will hide inside genuine holiday traffic.
- According to Salesforce, AI shopping agents drove an estimated 20% of 2025 holiday retail sales, about $262 billion, before Muse and Dots existed, and this season retailers will need to tell helpful agents from hostile ones in real time.
Preparing bot management for the holiday peak
If your holiday peak is weeks away, now is the time to pre-stage fingerprint-first bot management policies on promotions, sign-in, and checkout, before holiday bot attacks start. Request a live demo or contact us to talk through your launch or holiday readiness plan.
