This Week In Safety: Flock Cameras Are Outdated, Microsoft Patches Patches, And Researchers Attack SSH
Leak web site Distributed Denial of Secrets has released a dump of the filesystems of a Flock digicam, and Micah Lee has published a dive into the contents. Apparently the Flock safety mannequin didn’t embody “offended residents with a hacksaw in the course of the evening” within the bodily safety class.
The very first thing Micah notes is that the Flock {hardware} runs Android 8.1 (for these not retaining monitor, the present Android launch is Android 17 launched in June 2026). The model of Android working on the Flock digicam was final patched in June of 2018, and the Linux kernel (3.18.71) is over 9 years outdated, and the three.18 sequence went end-of-life in 2019.
You may assume “wouldn’t an working system this outdated have identified vulnerabilities?”, and also you’d be utterly proper. Micah calls out two specifically, one within the Qualcomm GPU which permits any program to control kernel reminiscence and achieve root (comparable, but less complicated, than the spate of kernel vulnerabilities this yr that allowed manipulating reminiscence by way of the disk IO cache), and the “WrongZone” vulnerability which permits a course of to escalate to root by way of socket dealing with errors. You may additionally guess each of those have been patched, and once more, you’d be proper, in 2021 and 2018 respectively.
Digging deeper, Micah discovers that API keys with entry into the Flock roads seem like hard-coded into the binaries. Each digicam seems to request credentials from an authentication server, utilizing the MAC deal with of the digicam. After acquiring the credentials from an Okta Auth0 sign-in service, the Flock shops them, in plain textual content.
Also saved on the digicam unencrypted are logs and place information: The digicam dumped to Distributed Denial of Secrets got here from a suburb of Milwaukee.
Wired additionally released research into the dumped filesystem photos. While some video and photograph information is encrypted on the system, the decryption keys are additionally saved on the system. The hacker collective who obtained the digicam decrypted the saved information, displaying that the digicam captured 1.6 million photographs of fifty,000 autos in underneath a month (21 days).
Despite claims from Flock that the detection solely captures autos, proof within the analyzed photographs and software program seems to point it intentionally captures individuals, and does so utilizing a video seize format as a substitute of the nonetheless photos used for license plates.
Flock, in the meantime, has mentioned that it’s not conscious of any safety points as a result of they haven’t been reported via the Flock safety web site, and can’t make a safety evaluation based mostly on incomplete reviews. Presumably identified vulnerabilities from the previous decade is just too troublesome?
Microsoft Releases emergency out-of-band fixes for Patch Tuesday fixes
Last week was a file breaking Patch Tuesday with almost 1000 safety fixes, and Microsoft has begun issuing recommendation that safety patches needs to be utilized instantly, or at most, inside three days. The threat of quickly making use of Microsoft patches in manufacturing or throughout massive fleets of machines is that Microsoft hasn’t had one of the best monitor file concerning patch stability, which brings us to this emergency set of patches to repair the patches.
The September 2026 safety replace precipitated points with distant desktop entry to servers, impacting the administration console, file explorer, and Windows replace instruments. It additionally precipitated folder sharing with Hyper-V digital machines to fail in some instances, and USB audio points with some units. The emergency patch additionally contains a number of safety fixes itself to repair privilege escalation bugs.
A handful of emergency fixes on high of a thousand patches could not really feel like many, however for company environments, shedding entry to distant desktop or entry to shared directories with digital machines may very well be a piece stopper, and any subject multiplied throughout probably hundreds of programs in a single firm instantly turns into a big value. If Microsoft expects company clients to have the ability to set up patches instantly, the testing and certification course of must enhance shortly. Every time the IT division or the CISO has to clarify downtime attributable to patches, it turns into tougher to deploy the subsequent set of patches.
Korea fights information breaches with fines
South Korea has raised the fines for major data breaches to as much as ten p.c of the businesses yearly proceeds.
Under the brand new rules, corporations with information leaks impacting ten million or extra individuals could be fined as much as ten p.c of the corporate yearly proceeds. Previously in South Korea, corporations concerned in massive information breaches may very well be fined as much as three p.c of yearly proceeds.
The fines are imposed on corporations which have had repeated violations inside the previous three years, and that are deemed to be grossly negligent. Not all violations routinely rise to the ten p.c restrict, and firms who can exhibit processes and financial backing in information safety can cut back the fines.
Shifting company conduct by way of legislative fines is troublesome, however could be the solely option to stem a number of the rampant points of information theft and extortion attributable to ransomware teams, and basing fines off yearly proceeds will certainly have a bigger impression than flat-rate fines which can have so little impression that it’s less complicated to pay them than resolve the issues.
RubyGems provide chain assault was truly OpenAI
In May of 2026, the RubyGems repository — the Ruby equal of NPM for Node or PyPI for Python — was hit with a wave of hundreds of malicious packages that tried to contaminate any customers who downloaded them as a part of a construct. Additionally, a code execution vulnerability was discovered within the RubyDoc doc generator, which allowed the attacker to execute arbitrary code within the dockerized occasion of RubyDocs used to generate documentation for uploaded packages. Between May 5 2026 and June 18 2026, over 2000 malicious packages have been uploaded.
At the time, the assault was labeled a “main malicious assault” however the functions have been unclear; extra recent research seems to point to this being a swarm of OpenAI brokers. The packages have been recognized as coming from LLM technology in the course of the preliminary assault, however it additionally seems that the brokers recognized themselves as OpenAI within the bundle commits, typically itemizing the writer as “oai”. This wouldn’t be sufficient to level at OpenAI definitively, anybody can declare an writer line, however in September, OpenAI has admitted to brokers breaking containment and listed information and assets they accessed. These correlate with the behaviors within the packages uploaded to RubyGems months beforehand.
Finally, as a part of the Gems submission course of, documentation is routinely generated by RubyDoc. The doc technology code had choices which might permit arbitrary command execution, which is generally disabled – however not within the docker containers used within the RubyDoc workflow. Due to this oversight, the RubyDocs web site has been susceptible to arbitrary execution from any supply, however it wasn’t till the agent swarm exploited it that it got here to mild. With full entry to the containers producing documentation, the brokers then tried to gather API keys of different customers, however the researchers at rubyhack.ai can’t verify that they really obtained entry.
Due to a caching subject with the CDN behind RubyGems, for as much as an hour after a person logged in utilizing a (legacy) model of Gems, the authentication token could be cached within the content material community. The LLM generated code tried to entry keys cached within the CDN, after which tried to push packages to the RubyGems repository utilizing the stolen keys. The brokers additionally exploited one other new vulnerability within the RubyGems system to bypass electronic mail authentication.
It will likely be fascinating to see if additional analysis identifies earlier incidences of OpenAI and different frontier brokers escaping containment and illegally hacking different corporations. [Editor’s Note: And exciting to see who is held responsible!]
NightmareEclipse reveals their id
The researcher often known as NightmareEclipse has been releasing exploits for Windows all through 2026, typically posting them simply after the Patch Tuesday cycle. Many of the releases have been accompanied by rants in opposition to Microsoft for ruining their lives. Combined with different feedback and knowledge leaks, it has been suspected that the researcher could have been a former Microsoft worker, and now it’s been confirmed directly.
Abdelhamid Naceri has confirmed they’re behind the NightmareEclipse id, and was beforehand credited with analysis by the Zero Day Initiative, a risk searching and backbone group. Naceri claims to have been illegally fired by Microsoft, and has been concerned in a number of courtroom instances in Germany preventing his termination, whereas in courtroom Microsoft refused to supply particulars about vulnerabilities Naceri was concerned in that will have led to his termination.
Revolut phished
The mobile money firm Revolut was phished with spoofed government domains, revealing the small print of an unknown variety of clients.
Revolut operates as a financial institution in lots of corporations and claims over 80 million clients worldwide, and is concerned in cryptocurrency buying and selling. Using an undisclosed compromised authorities company area, attackers requested buyer particulars, and Revolut supplied. The information included buyer addresses, birthdates, id documentation equivalent to passports and drivers licenses, and in some instances transaction historical past and different particulars.
It’s unclear what number of clients have been impacted, or why this info was out there to an electronic mail request from a authorities company with out extra validation. Given the high-profile clients Revolut courts, this may increasingly have been a focused assault in opposition to a high-value group of shoppers to help with future phishing makes an attempt.
Attacks in opposition to SSH streams
A paper from the ACM CCS 2026 convention particulars assaults in opposition to OpenSSH by manipulating and analyzing the stream throughout compression.
SSH permits a number of channels on one connection; along with the traditional interactive shell, SSH connections can carry out port forwarding, act as SOCKS proxies, and even act as a full VPN with digital community interfaces. OpenSSH connections additionally allow compression, however the compression happens throughout the information from all channels concurrently.
The paper exhibits that if an attacker is ready to manipulate the information in a single channel, they are able to affect the compression and derive the contents of information in different channels on the identical connection. When SSH is used for port forwarding, for instance, the forwarded server is probably not trusted, however can nonetheless affect the packets generated. The accuracy and effectivity of the assault could be influenced by how a lot exercise happens on different channels of the connection, however the researchers exhibit having the ability to get well an 8 character secret utilizing a 26 letter alphabet (like a primary password) in 276 guesses.
SSH permits forwarding of terminal classes, X11 unix sockets, native Unix area sockets, TCP sockets, and straight linked file descriptor channels; whereas every channel is maintained individually, as soon as the information is mixed and is prepared for transmit, all of the channels are compressed on the similar time. At essentially the most primary stage, compression algorithms search for repeating information that may be lowered to a single file, and the researchers present that by influencing the information being compressed, and analyzing the encrypted, compressed information generated, guesses could be made in regards to the content material of different channels.
The paper exhibits assaults in opposition to Ansible secrets and techniques, browser plaintext information, and easy plaintext auth utilized by Redis. Currently the assault mannequin is comparatively esoteric; possible there isn’t a rapid risk to nearly all of SSH customers. Similar assaults have been demonstrated in opposition to the TLS protocol, ensuing within the elimination of compression as a part of the usual. If no different options could be discovered, it’s cheap to suppose that compression could also be faraway from SSH shoppers and servers as a mitigation in opposition to these kinds of assaults.
