CarPlay Exploit, BYOVD Techniques, SQL C2 Attacks, iCloud Backdoor Demand & Extra

Anthropic mentioned it has rolled out quite a lot of security and security enhancements to Claude Sonnet 4.5, its newest coding centered mannequin, that make it tough for unhealthy actors to use and safe the system towards immediate injection assaults,
sycophancy (i.e., the tendency of an AI to echo and validate consumer beliefs irrespective of how delusional or dangerous they could be),
and little one security dangers. “Claude’s improved capabilities and our intensive security coaching have allowed us to considerably enhance the mannequin’s conduct, decreasing regarding behaviors like sycophancy, deception, power-seeking, and the tendency to encourage delusional pondering,”
the corporate mentioned. “For the mannequin’s agentic and laptop use capabilities, we have additionally made appreciable progress on defending towards immediate injection assaults, one of the crucial critical dangers for customers of those capabilities.”
The AI firm mentioned the newest mannequin has higher defensive cybersecurity skills, resembling vulnerability discovery, patching, and fundamental penetration testing capabilities. Nevertheless, it did acknowledge that these instruments could possibly be “dual-use,” which means they may additionally probably be utilized by malicious actors, in addition to cybersecurity professionals.
Generative AI programs like these supplied by Microsoft and OpenAI are on the forefront of a battle between corporations offering refined textual content and picture technology capabilities and malicious actors seeking to exploit them.

The SANS Web Storm Heart Safety has disclosed its commentary of a big improve in internet-wide scans concentrating on the important PAN-OS GlobalProtect vulnerability
(CVE-2024-3400). The vulnerability, disclosed final 12 months, is a command injection vulnerability that could possibly be exploited by an unauthenticated attacker to execute arbitrary code with root privileges on prone firewalls.
SANS ISC mentioned it has detected specifically crafted requests that search to add a TXT file and subsequently try to retrieve that file by way of an HTTP GET request. “This can return a ‘403’ error if the file exists, and a ‘404’ error if the add failed. It won’t execute code,” it
famous. “The content material of the file is a regular International Shield session file, and won’t execute. A follow-up assault would add the file to a location that results in code execution.”
In current weeks, exploit makes an attempt have additionally been registered towards Hikvision cameras prone to an older flaw
(CVE-2017-7921), SANS ISC mentioned.

A classy assault marketing campaign has focused improperly managed Microsoft SQL servers to deploy the open-source
XiebroC2 command-and-control (C2) framework utilizing PowerShell to determine persistent entry to compromised programs.
The assault leverages susceptible credentials on publicly accessible database servers, permitting risk actors to acquire an preliminary foothold and escalate privileges by a device known as JuicyPotato.
“XiebroC2 is a C2 framework with open-source code that helps numerous options resembling info assortment, distant management, and protection evasion, just like Cobalt Strike,” AhnLab
mentioned.

Google has outlined the varied hardening suggestions that organizations can take to safeguard towards assaults mounted by
UNC6040, a financially motivated risk cluster that makes a speciality of voice phishing (vishing) campaigns particularly designed to compromise organizations’ Salesforce cases for large-scale information theft and subsequent extortion.
Central to the operation entails deceiving victims into authorizing a malicious related app to their group’s Salesforce portal.
“Over the previous a number of months, UNC6040 has demonstrated repeated success in breaching networks by having its operators impersonate IT assist personnel in convincing telephone-based social engineering engagements,” it
mentioned.
“This method has confirmed significantly efficient in tricking workers, usually inside English-speaking branches of multinational firms, into actions that grant the attackers entry or result in the sharing of delicate credentials, finally facilitating the theft of the group’s Salesforce information. In all noticed instances, attackers relied on manipulating finish customers, not exploiting any vulnerability inherent to Salesforce.”

Censys mentioned it recognized over 60 cryptocurrency phishing pages impersonating well-liked {hardware} pockets manufacturers Trezor and Ledger by an evaluation of robots.txt information.
These websites have an entry within the file: “Disallow: /add_web_phish.php.”
“Notably, the actor behind the pages tried to dam well-liked phishing reporting websites from indexing the pages by together with endpoints of the phishing reporting websites in their very own robots.txt file,” the corporate mentioned.
The weird robots.txt sample has additionally been found on a number of GitHub repositories, a few of which date again to January 2025.
“The misuse of robots.txt and the merge conflicts present in a number of READMEs might additionally counsel that the actor behind these pages will not be well-versed in internet improvement practices,” security researcher Emily Austin
added.

Google has introduced that it is updating Google Drive for desktop with AI-powered ransomware detection to robotically cease file syncing and permit customers to simply restore information with a couple of clicks.
“Our AI-powered detection in Drive for desktop identifies the core signature of a ransomware assault — an try to encrypt or corrupt information en masse — and quickly intervenes to place a protecting bubble round a consumer’s information by stopping file syncing to the cloud earlier than the ransomware can unfold,” Google Cloud
mentioned.
“The detection engine adapts to novel ransomware by repeatedly analyzing file adjustments and incorporating new risk intelligence from VirusTotal. When Drive detects uncommon exercise that implies a ransomware assault, it robotically pauses syncing of affected information, serving to to forestall widespread information corruption throughout a company’s Drive and the disruption of labor.”
Customers subsequently obtain an alert on their desktop and by way of e-mail, guiding them to revive their information. The actual-time ransomware detection functionality is constructed atop a specialised AI mannequin skilled on hundreds of thousands of actual sufferer information encrypted by numerous ransomware strains.

Imgur, a well-liked picture internet hosting platform with greater than 130 million customers, has blocked entry to customers within the U.Okay. after regulators signalled their intention to impose penalties over considerations round kids’s information.
The U.Okay.’s information watchdog, the Info Commissioner’s Workplace (ICO), mentioned it lately notified the platform’s dad or mum firm, MediaLab AI, of plans to effective Imgur after investigating its method to age checks and dealing with of youngsters’s private information.
The probe was launched earlier this March.
“Imgur’s resolution to limit entry within the U.Okay. is a business resolution taken by the corporate,” the ICO
mentioned.
“We’ve been clear that exiting the U.Okay. doesn’t permit an organisation to keep away from accountability for any prior infringement of knowledge safety legislation, and our investigation stays ongoing.”
In a assist web page, Imgur
confirmed U.Okay. customers will be unable to log in, view content material, or add pictures.

An audit of the Russian authorities’s new MAX immediate messenger cellular app has discovered no proof of surveillance past accessing options vital for the app to perform.
“Throughout two days of commentary, no take a look at configurations revealed improper entry to the digicam, location, microphone, notifications, contacts, images, and movies,” RKS International
mentioned.
“Technically, the applying had the power to gather these information and ship them, however specialists didn’t report what occurred. After revoking permits, the applying doesn’t report makes an attempt to acquire these accesses once more by requests or unauthorized.”

The U.Okay. authorities has issued a brand new request for Apple to supply entry to encrypted iCloud consumer information, this time focusing particularly on the ‌iCloud‌ information of British residents, in accordance with the Monetary Occasions.
The request, issued in early September 2025, has demanded that Apple create a manner for officers to entry encrypted ‌iCloud‌ backups.
In February, Apple withdrew ‌iCloud‌’s Superior Data Safety characteristic within the U.Okay.
Subsequent pushback from civil liberty teams and the U.S. authorities led to the U.Okay. apparently abandoning its plans to power Apple to weaken encryption protections and embody a backdoor that might have enabled entry to the protected information of U.S. residents.
In late August, the Monetary Occasions additionally reported that the U.Okay. authorities’s secret order was “not restricted to” Apple’s ADP characteristic and included necessities for Apple to “present and keep a functionality to reveal classes of knowledge saved inside a cloud-based backup service,” suggesting that the entry was far broader in scope than beforehand identified.

Again in April 2025, Oligo Safety disclosed a set of flaws in AirPlay known as
AirBorne (CVE-2025-24252 and CVE-2025-24132) that could possibly be chained collectively to take over Apple CarPlay, in some instances, with out even requiring any consumer interplay or authentication.
Whereas the underlying expertise makes use of the iAP2 protocol to determine a wi-fi connection over Bluetooth and negotiate a CarPlay Wi-Fi password to permit an iPhone to connect with the community and provoke display screen mirroring, the researcher discovered that many units and programs default to a “No-PIN” method through the Bluetooth pairing part, making the assaults “frictionless and tougher to detect.”
This, coupled with the truth that iAP2 doesn’t authenticate the iPhone, meant that an attacker with a Bluetooth radio and a suitable iAP2 shopper can impersonate an iPhone, request the Wi-Fi credentials, set off app launches, and situation any arbitrary iAP2 command.
From there, attackers can exploit CVE-2025-24132 to realize distant code execution with root privileges.
“Though patches for CVE-2025-24132 have been revealed on April 29, 2025, just a few choose distributors really patched,” Oligo
mentioned.
“To our data, as of this submit, no automotive producer has utilized the patch.”

Russia’s Ministry of Digital Growth is engaged on laws to power corporations to limit the kind of information they acquire from residents within the nation, within the hopes of minimizing future leaks of confidential information.
“Techniques shouldn’t course of info containing private information past what is critical to make sure enterprise processes,”
mentioned Evgeny Khasin, appearing director of the Ministry of Digital Growth’s cybersecurity division.
“It’s because many organizations have a tendency to gather as a lot information as attainable in an effort to work together with it in a roundabout way or use it for their very own functions, whereas the legislation stipulates that information ought to be minimized.”

The Dutch authorities has mentioned it will not assist Denmark’s proposal for an E.U. Chat Management laws to power tech corporations to introduce encryption backdoors in order to scan communications for “abusive materials.”
The proposal is up for a vote on October 14.
The Digital Frontier Basis (EFF) has known as the legislative proposal “harmful” and tantamount to “chat surveillance.”
Different E.U. nations which have opposed the controversial laws embody Austria, Czechia, Estonia, Finland, Luxembourg, and Poland.

Google has agreed to pay $48 million, and the menstrual monitoring app Flo Well being can pay $8 million to resolve a category motion lawsuit alleging the app illegally shared folks’s well being information.
Google is anticipated to arrange a $48 million fund for Flo app customers who entered details about menstruation or being pregnant from November 2016 till the tip of February 2019.
In March 2025, defunct information analytics firm Flurry mentioned it could pay $3.5 million for harvesting sexual and reproductive well being information from the interval monitoring app.
The criticism, filed in 2021, alleged that Flo used software program improvement kits to permit Google, Meta, and Flurry to intercept customers’ communications inside the app.

Meta Platforms mentioned it plans to start out utilizing folks’s conversations with its AI chatbot to assist personalize adverts and content material.
The coverage is ready to enter impact on December 16, 2025. It will not apply to customers within the U.Okay., South Korea, and the European Union, for now.
Whereas there isn’t a opt-out mechanism, conversations associated to spiritual or political opinions, sexual orientation, well being, and racial or ethnic origin might be
robotically excluded from the corporate’s personalization efforts.
The corporate mentioned its AI digital assistant now has greater than 1 billion energetic month-to-month customers.

The Federal Commerce Fee (FTC) has sued Sendit’s working firm, Iconic Hearts, and its CEO for “unlawfully amassing private information from kids, deceptive customers by sending messages from faux ‘folks,’ and tricking shoppers into buying paid subscriptions by falsely promising to disclose the senders of nameless messages.”
The company mentioned,
“Despite the fact that it was conscious that many customers have been underneath 13, Iconic Hearts didn’t notify mother and father that it collected private info from kids, together with their telephone numbers, birthdates, images, and usernames for Snapchat, Instagram, TikTok, and different accounts, and didn’t acquire mother and father’ verifiable consent to such information assortment.”

Risk actors are promoting entry to MatrixPDF, a device that lets them alter atypical PDF information to lures that may redirect customers to malware or phishing websites.
“It bundles phishing and malware options right into a builder that alters legit PDF information with faux safe doc prompts, embedded JavaScript actions, content material blurring, and redirects,” Varonis
mentioned.
“To the recipient, the file seems routine, but opening it and following a immediate or hyperlink can lead to credential theft or payload supply.”

Microsoft mentioned it is planning to introduce a brand new Edge security characteristic that may shield customers towards malicious extensions sideloaded into the online browser.
“Microsoft Edge will detect and revoke malicious sideloaded extensions,” it mentioned.
The rollout is anticipated to start out someday in November 2025. It didn’t present additional particulars on how these harmful extensions might be recognized.

The U.S. authorities prolonged the deadline for ByteDance to divest TikTok’s U.S. operations till December 16, 2025, making it the fourth such extension.
The event got here as China mentioned the U.S. spin-off of TikTok will use ByteDance’s Chinese language algorithm as a part of a U.S.-agreed framework that features “licensing the algorithm and different mental property rights.”
The bogus intelligence (AI)-powered algorithm that underpins the app has been a supply of concern amongst nationwide security circles, because it could possibly be manipulated to push Chinese language propaganda or polarizing materials to customers.
China has additionally known as the framework deal a “win-win.”
Below the framework deal, about 80% of TikTok’s U.S. enterprise could be owned by a three way partnership that features Oracle, Silver Lake Companions, media mogul Rupert Murdoch, and Dell CEO Michael Dell, with ByteDance’s stake dropping beneath 20% to adjust to the nationwide security legislation.
The divestiture additionally extends to different functions like Lemon8 and CapCut which can be operated by ByteDance.
Moreover, TikTok’s algorithm might be copied and retrained utilizing U.S. consumer information as a part of the deal, with Oracle auditing the advice system.
The White Home has additionally promised that every one U.S. consumer information on TikTok might be saved on Oracle servers within the U.S.

An info stealer generally known as Acreed is gaining traction amongst risk actors, with a gentle rise in Acreed logs in Russian-speaking boards. The stealer was first marketed on the Russian Market in February 2025 by a consumer named “Nu####ez” and is assessed to be a personal mission. As of September 2025, the highest 5 info stealer strains included Rhadamanthys (33%), Lumma (33%), Acreed (17%), Vidar (12%), and StealC (5%). “At the moment, Acreed is possibly a privately developed mission, however our infrastructure evaluation reveals that additionally it is built-in in an current ecosystem that overlaps with Vidar,” Intrinsec
mentioned.

Cybersecurity firm Sophos mentioned it noticed Warlock ransomware actors (aka Storm-2603 or Gold Salem) abusing the legit open-source
Velociraptor digital forensics and incident response (DFIR) device to determine a Visible Studio Code community tunnel inside the compromised surroundings. A few of the incidents led to the deployment of the ransomware. Warlock
gained
prominence in July 2025 after it was discovered to be of the risk actors abusing a set of security flaws in Microsoft SharePoint known as ToolShell to infiltrate goal networks. The group has claimed 60 victims as of mid-September 2025, beginning its operations in March, together with a Russian firm, suggesting that it could be working from exterior the Kremlin. Microsoft has described it with reasonable confidence as a China-based risk actor. The group has additionally been noticed weaponizing the ToolShell flaws to drop an ASPX internet shell that is used to obtain a Golang-based WebSockets server that enables continued entry to the compromised server independently of the online shell. Moreover, Gold Salem has employed the Carry Your Personal Weak Driver (BYOVD) method to bypass security defenses through the use of a vulnerability (CVE-2024-51324) within the Baidu Antivirus driver BdApiUtil.sys to terminate EDR software program. “The rising group demonstrates competent tradecraft utilizing a well-known ransomware playbook and hints of ingenuity,” Sophos
mentioned.

Risk actors are distributing faux Chrome extensions posing as synthetic intelligence (AI) instruments like OpenAI ChatGPT, Llama, Perplexity, and Claude. As soon as put in, the extensions let customers kind prompts within the Chrome search bar, however will hijack the prompts to redirect queries to attacker-controlled domains and observe search exercise. The browser add-ons “override the default search engine settings by way of the chrome_settings_overrides manifest key,” Palo Alto Networks Unit 42
mentioned. The queries are redirected to domains like chatgptforchrome[.]com, dinershtein[.]com, and gen-ai-search[.]com.

A classy operation has been discovered to interrupt into routers and IoT units utilizing weak credentials and identified security flaws, and hire the compromised units to different botnet operators.
The operation has witnessed a significant spike in exercise this 12 months, leaping 230% in mid-2025, with the botnet loader-as-a-service infrastructure used to ship payloads for DDoS and cryptomining botnets like RondoDoX, Mirai, and Morte, per
CloudSEK.

Tile location trackers leak delicate info that may permit risk actors to trace a tool’s location. That is in accordance with researchers from the Georgia Institute of Know-how, who reverse-engineered the location-tracking service and located that the units leak MAC addresses and distinctive gadget IDs.
An attacker can reap the benefits of the absence of encryption protections to intercept and acquire the data utilizing a easy radio antenna, finally enabling them to trace the entire firm’s prospects.
“Tile’s servers can persistently be taught the placement of all customers and tags, unprivileged adversaries can observe customers by Bluetooth ads emitted by Tile’s units, and Tile’s anti-theft mode is definitely subverted,” the researchers
mentioned in a examine.
The problems have been reported to its dad or mum firm Life360 in November 2024, following which it mentioned a
“variety of enhancements” have been rolled out to handle the issue, with out specifying what these have been.

New statistics launched by Forescout present {that a} quarter of all OpenSSH and eight.5% of all SSH servers now assist post-quantum cryptography (PQC).
In distinction, TLSv1.3 adoption stays at 19% and TLSv1.2 – which doesn’t assist PQC – elevated from 43% to 46%.
The report additionally discovered that manufacturing, oil and fuel, and mining have the bottom PQC adoption charges, whereas skilled and enterprise companies have the very best.
“Absolutely the variety of servers with PQC assist grew from 11.5 million in April to virtually 15 million in August, a rise of 30%,” it
added.
The relative quantity grew from 6.2% of whole servers to eight.5%.

Synacktiv has documented a brand new method to programmatically inject and activate Chrome extensions in Chromium-based browsers inside Home windows domains for malicious functions by manipulating Chromium inside desire information and their related JSON MAC property (“super_mac”).
The analysis “highlights the inherent problem in cryptographically defending browser-internal secrets and techniques just like the MAC seed, as any really strong answer would want to account for various working system-specific security mechanisms (like DPAPI on Home windows) with out affecting cross-platform compatibility,” the corporate
mentioned.

An e-mail phishing marketing campaign has been noticed concentrating on entities within the greater training sector to steal credentials and Cisco Duo one-time passwords (OTPs) with the aim of compromising accounts, exfiltrating information, and launching lateral assaults.
“Targets are funneled to spoofed sign-in portals that completely mimic college login pages,” Irregular AI
mentioned.
“Then, purpose-built phishing kits harvest each credentials and Duo one-time passwords (OTPs) by seamless multi-step flows. With these particulars in hand, attackers swiftly hijack accounts, conceal their tracks with malicious mailbox guidelines, and launch lateral phishing campaigns inside the similar group.”
Greater than 40 compromised organizations and over 30 focused universities and schools have been recognized as a part of the marketing campaign.