Latest

Cybersecurity

Dozens of Broadcom routers remotely vulnerable since 2010

Two flaws in Broadcom's proprietary Wi-Fi allow code execution as root without the network password. The 2018 patch never reached the installed base.

Dozens of Broadcom routers vulnerable since 2010
Dozens of Broadcom routers vulnerable since 2010

Quentin Kaiser, a security researcher, presented two flaws in the parser of Broadcom's proprietary 802.11 implementation at the Hexacon conference in Paris and published the full technical analysisthe same day. With them, anyone within radio range can execute code as root without knowing the network password and without completing any pairing.

The flaw is neither new nor alien to the industry. Broadcom released a silent patch in 2018 that covered only part of the problem and then changed course: the next version of its SDK dropped the affected component and replaced it with hostapd, the open daemon used in Linux distributions, with a recommendation to migrate. Those who did not migrate are still running the old code, and on devices running eCos, a minimal operating system, there is nowhere to migrate to.

No one has the exact scope, and the figures in circulation are the researcher's own estimates: dozens of manufacturers, hundreds of products and millions of access points, with the attack demonstrated on home gateways from three of the four major Belgian operators. The difference from an ordinary CVE lies in the way out, because here there is no new version to install across much of the installed base, but rather an operator decision that almost no one has taken.

The overflow starts with a greeting

WPS exists so that no one has to type a password on a television. You press a button or enter an eight-digit PIN, and the client and the access point exchange EAP messages until they agree on a key before the former is authenticated.

Those messages allow fragmentation, so the sender splits its own and the access point reassembles them. Broadcom wrote that reassembly by hand in a daemon called wps_monitor, inside its proprietary 802.11 implementation. Kaiser found two flaws in the validation of fragment length: the code trusts the declared size and copies more bytes into an internal buffer than fit.

The result is memory disclosure and controlled writing, and with those two pieces code execution is assembled. The daemon runs as root, so the attacker inherits those privileges. There is no need for the network key, nor to complete a normal WPS pairing, nor for anyone to press a button: it is enough for the device to advertise WPS and be within antenna range.

The proximity seen in the demonstration videos is down to the laboratory setup, as the researcher himself clarifies. In practice, radio range describes a street, a building entrance or the room next door in a hotel.

A flaw that lives in the SDK, not in the router

The firmware of a home router is almost never its own code. The manufacturer assembles an image on top of the SDK it buys, adds its web interface and its configuration layer, and signs the whole thing. The wireless stack arrives compiled by Broadcom, so the manufacturer neither sees the lines where the error is nor can rewrite them.

Broadcom released an incomplete patch in 2018 that did not close the hole and then shifted the problem to its customers, who had to adapt their software. That is where a security update becomes an engineering project that someone has to fund and validate with their operator.

SDK platform Affected chipsets Remaining leeway
eCos BCM3380, BCM3383 and BCM33834 None, because the new SDK does not run there and the flaw stays forever
Linux BCM3390, BCM63xx, BCM65xx and BCM68xx Depends on the manufacturer migrating to hostapd and releasing firmware

eCos is a minimal operating system designed for devices with very little memory, and on that branch the flaw is irreversible by design. On the Linux branch the door remains open, but each manufacturer is the one who has to walk through it, and no one has published a timetable.

What Kaiser achieved in Belgium

On the ASKEY TCG3000 gateway that Orange Belgium provides to its customers, his exploit runs code that reads the Wi-Fi key from NVRAM memory and returns it to the attacker in 802.11 frames. It then connects with that same key to prove it was the right one. On the Sagemcom F@st that Proximus distributes as the BBOX 3V+, he opened a reverse shell using only radio frames, with no command server in the cloud.

The devices named are Netgear, Technicolor and Sagemcom models that the Belgian operators have been distributing for years. The researcher's figures, based on subscriber data and medium confidence, put the scope above one and a half million gateways in a single country, and they are not a verified inventory.

Operator Customers Estimated vulnerable gateways Situation as of October 20
Proximus 2.318.000 850.000 WPS disabled across the entire fleet, patch promised within weeks
VOO (legacy fleet) about 500,000 500.000 No public measures
Orange Belgium about 547,000 210.000 Considering turning off WPS across the entire infrastructure
Telenet 1.758.000 no figure Out, according to its own check

The reaction came in stages. Telenet checked its entire fleet for the vulnerable daemon and did not find it, so it is out. Proximus disabled WPS on all its BBOX devices within 24 hours and promised the full patch in the following weeks, and Kaiser considered the case resolved for now. Orange Belgium was still studying a mass shutdown when the update was published.

A CVE that never existed

The timeline of the process explains why the flaw is still alive better than any analysis of the code. Kaiser sent the first report to Broadcom on July 18, 2026, and the second nine days later, both with a 90-day deadline. Broadcom replied on July 28 that a patch existed and that he should notify the manufacturers so they could contact them.

The next day, the company refused to register a CVE and declined to share the notification it had sent to manufacturers through its customer portal. CERT/CC opened a case on August 4 and closed it the same day, because Broadcom is a CVE numbering authority and did not intend to assign any. Without a CVE there is no alert to cross-check against an inventory.

The operators received the report on July 29 and had 78 days until the talk. On October 15, a day before Hexacon, none of the devices tested in Belgium had received the patch.

When the warning exists and circulates, the chain works: the intermediates of the Indian CA eMudhra, with keys anyone could recalculate, were blocked by Chrome and Firefox within hours, as we reported when reviewing the compromised TLS keys. Here there was neither a warning nor a CVE, something serious that points directly at Broadcom and a poor internal policy for handling these cases.

What can someone with one of these devices do?

The first step is to find out whether the device advertises WPS. Kaiser published a YARA rule and a fingerprinting script that, with a card in monitor mode, distinguishes the Broadcom stack from hostapd: it sends a fragmented frame of zero length and checks whether the device responds with ACK or NACK. If the access point does not advertise WPS, the tool declares itself inconclusive.

Disabling WPS is what Proximus did within hours and the researcher accepted it as sufficient mitigation. PCWorld argues instead that turning off the feature does not close the hole on all models, and no one has published a model-by-model test. While that is being clarified, the prudent course is to treat the operator's router network as a public network.

Changing the firmware is not a general way out either, because on devices with a Broadcom wireless chip the community covers only a small part of the catalogue, since the manufacturer's own driver is missing.

What is within reach is to place your own access point behind it, separate the household devices onto a different network and not leave administration exposed to the internet. That compartmentalization does not fix the router, but it limits what an attacker finds if they get in, and it works the same with or without an operator router.

The slice of the installed base that no one updates

The lesson is not in the Broadcom flaw, which is one among thousands, but in where it lives. Home network hardware is the only piece of the installed base that no one updates: it has no app store, no reliable automatic updates, and no user who checks the firmware version. It is installed with the contract and stays there for a decade.

Kaiser found two flaws in less than a week using frontier AI models, and he reported it as good news because it makes auditing the deep layers of infrastructure cheaper. The same cost reduction serves whoever is looking for a crack in the neighbour's network, and nearest-neighbour attacks against hotel networks already appear in incident response teams' reports. The uncomfortable question is not how many routers are affected, but how many will receive an update before they are retired.

José Maldonado

Author

José Maldonado

Activist and blogger on technology, free software and blockchain.

The latest
From Huang to Botín, from Zara to Byron: corporate responsibility in global capitalism

Keep reading