Sale of 0-day exploits: macOS LPE and several pre-auth/post-auth RCE vulnerabilities

For informational purposes only.

1. 0-day LPE for macOS

Vulnerability type: Local Privilege Escalation

According to [StatCounter](https://gs.statcounter.com/os-market-share/desktop), macOS accounts for approximately 15% of desktop operating systems worldwide, making it the second most popular desktop OS after Windows. According to [JumpCloud SME IT Trends](https://www.computerworld.com/article/3488606/jumpcloud-survey-highlights-the-business-case-for-macs.html), on average, the IT infrastructure of small and medium-sized businesses includes about 24% macOS devices (worldwide).

2. Pre-auth RCE for PaperCut

Vulnerability type: Remote Code Execution without authentication

According to the vendor's own data, [PaperCut](https://www.papercut.com/) is installed in [more than 70,000 organizations in 195 countries](https://www.papercut.com/), with a total of over 100 million users – including educational institutions, corporations, and government agencies worldwide.

It is worth noting that PaperCut has already been a real, not hypothetical, target for hacker groups. In the ["Current Cyber Threats: Q2 2023"](https://global.ptsecurity.com/analytics/cybersecurity-threatscape-2023-q2) report, Positive Technologies classifies [CVE-2023-27350](https://dbugs.ptsecurity.com/vulnerability/CVE-2023-27350) and [CVE-2023-27351](https://dbugs.ptsecurity.com/vulnerability/CVE-2023-27351) as critically dangerous – these vulnerabilities were actively exploited by ransomware groups Bl00dy and Cl0p against educational and government institutions.

3. Vulnerability in F5 BIG-IP

Vulnerability type: Remote Code Execution without authentication

[BIG-IP](https://www.f5.com/products/big-ip-services) is one of the most widely used products for load balancing and WAF in the world, installed in major banks, telecommunications companies, and corporations. According to [Shadowserver Foundation](https://www.bleepingcomputer.com/news/security/over-14-000-f5-big-ip-apm-instances-still-exposed-to-rce-attacks/), after the disclosure in October 2025 of the F5 internal network breach (attributed to China), the number of BIG-IP instances detected on the internet exceeded 266,000 IP addresses – over 142,000 in the United States, and approximately 100,000 in Europe and Asia. F5 serves more than 23,000 customers worldwide, including 48 companies from the Fortune 50 list.

4. Vulnerability in Adobe ColdFusion

Vulnerability type: Remote Code Execution without authentication

[ColdFusion](https://www.adobe.com/products/coldfusion-family.html) remains a niche but stable player in the enterprise web application segment on the Java stack.

This post details zero-day exploit sales including macOS privilege escalation and unauthenticated remote code execution vulnerabilities in PaperCut, F5 BIG-IP, and Adobe ColdFusion software affecting organizations globally.

Sale of 0-day exploits: macOS LPE and several pre-auth/post-auth RCE vulnerabilities

SensePost researchers have demonstrated a new way to inject code into another process without using the usual `WriteProcessMemory` or `VirtualAllocEx` functions that EDR solutions routinely monitor. The technique, dubbed Process Parameter Poisoning, manipulates process startup parameters (`lpCommandLine`, `lpEnvironment`, `lpReserved` from `STARTUPINFOW`), which are copied during process creation into the `RTL_USER_PROCESS_PARAMETERS` structure inside the `PEB` of the new process.

This approach lets malicious code execute in the target process without traditional injection artifacts, completely bypassing system call monitoring. In testing, the technique successfully evaded four popular EDR solutions, achieving code execution within the context of the new process.

Article: https://sensepost.com/blog/2026/process-parameter-poisoning/

Tool: https://github.com/Orange-Cyberdefense/p3-loader/

SensePost researchers demonstrated Process Parameter Poisoning, injecting code via startup parameters to bypass EDR monitoring. This technique avoids traditional artifacts, successfully evading four popular security solutions during testing.

Process Parameter Poisoning. Stealthy injection through process parameters

Conference recordings from BlueHat IL 2026 are now online! 

[BlueHat IL 2026 conference page](https://www.microsoftrnd.co.il/bluehatil/conference/about)

[Full conference agenda](https://www.microsoftrnd.co.il/bluehatil/conference/schedule)

[Keynote video recordings](https://www.youtube.com/playlist?list=PLH3m7nQBwGZo)

We've highlighted a few talks worth checking out:

• [Recovery Reimagined: Attacking and Securing Windows Recovery](https://www.youtube.com/watch?v=DkuEmspDX1s). The researchers present their study of Windows Recovery Environment and 11 new vulnerabilities that allow bypassing BitLocker and extracting data from encrypted drives.

• [ShadowMQ: Leveraging AI Architecture for RCE Across Leading Inference Servers](https://www.youtube.com/watch?v=15xqr43oCko). The speaker discusses remote code execution flaws in internal channels of popular AI inference servers, including projects from Meta, NVIDIA, Microsoft, vLLM, and SGLang.

• [The Stream Is Dead, Long Live the Stream: How HTTP/2 Lets Dead Streams Keep Servers Working](https://www.youtube.com/watch?v=GdSopFCPvIE). The talk covers the MadeYouReset vulnerability in HTTP/2 that enables request floods and denial of service. The issue affected Tomcat, Netty, F5, and several other implementations.

• [Exploiting AI Orchestration Zero-Days via PostgreSQL Internals](https://www.youtube.com/watch?v=71QAblR4d1Q). The researchers disclose multiple 0-day vulnerabilities in Langflow, n8n, and Activepieces — including remote code execution, sandbox escape, session hijacking, and arbitrary file read via PostgreSQL internals.

• [ChainLeak: From AI Framework to Cloud Secrets](https://www.youtube.com/watch?v=DYMKhDvq9hQ). This talk examines two vulnerabilities in Chainlit that enable remote file theft, cloud key and secret exfiltration, SSRF exploitation, and eventual server compromise.

• [Hybrid Cloud – A Novel Vulnerability Playground](https://www.youtube.com/watch?v=Cm8keeVjMjY). The presenters analyze four remote code execution and privilege escalation vulnerabilities in Azure Arc extensions that link on-prem systems with the cloud.

• [VoidLink Internals Inside Chinese Commercial Grade Cloud Native Malware](https://www.youtube.com/watch?v=ZJeSwZucI6A). The researchers dissect the VoidLink cloud-based malware framework, which integrates Linux kernel modules, eBPF, cloud credential theft, and over 30 components for offensive operations.

Microsoft recently released BlueHat IL 2026 recordings featuring vulnerabilities in Windows Recovery, AI servers, HTTP/2, AI orchestration tools, Chainlit, Azure Arc, and Chinese cloud malware VoidLink.

Conference recordings from BlueHat IL 2026 are now online!

Exploit for Sale: TeamViewer

For informational purposes only.

Vulnerability Type: Server-Side Request Forgery (SSRF)

The seller is offering an exploit for [TeamViewer](https://www.teamviewer.com/). This software is one of the most widely used remote access solutions in the world. According to the seller's description, the SSRF vulnerability allows the TeamViewer server-side component to be tricked into making a request to an internal cloud metadata endpoint, thereby obtaining temporary credentials for the cloud instance on which the infrastructure is deployed.

The author explicitly states that this is not a full Remote Code Execution (RCE), as the vulnerability itself does not immediately allow for code execution. However, by obtaining the cloud instance credentials, an attacker with sufficient time can further escalate privileges within the cloud environment, potentially leading to complete control of the instance. A similar chain of events was at the root of several high-profile incidents, [including the attack on Capital One](https://krebsonsecurity.com/2019/08/what-we-can-learn-from-the-capital-one-hack/) in 2019, where an SSRF vulnerability, combined with misconfigured IAM access, resulted in the leakage of data for over 100 million customers.

TeamViewer is one of the most popular remote access products in the world: according to [Wikipedia](https://en.wikipedia.org/wiki/TeamViewer), the application is installed on more than 2.5 billion devices, supports remote administration of equipment from over 130 manufacturers, and is actively used by both individual users and corporate IT departments. 

Positive Technologies has conducted research on the use of TeamViewer as a legitimate channel by attackers. In a [report summarizing the results of 2024](https://www.itsec.ru/news/positive-technologies-predstavila-pg-ngfw-dlia-zashiti-biznesa-ot-kiberatak), it was noted that 30% of malware used in successful attacks on Russian organizations are remote management tools, "perfectly legitimate in organizations, such as AnyDesk and TeamViewer"; the use of legitimate software to conceal activity is one of the key trends of the past two years.

A TeamViewer SSRF exploit is for sale, enabling cloud credential theft via metadata endpoints. Attackers escalate privileges, resembling the Capital One breach where SSRF and misconfigured access caused leakage.

Exploit for Sale: TeamViewer

CVE-2026-32202: The Trust Check Runs at the Click. The Coercion Runs at the Icon

Akamai researchers analyzed the Windows Shell vulnerability [CVE-2026-32202](https://dbugs.ptsecurity.com/vulnerability/CVE-2026-32202), which resulted from an incomplete fix for the previously exploited [CVE-2026-21510](https://dbugs.ptsecurity.com/vulnerability/CVE-2026-21510).

In February 2026, Microsoft patched the remote code execution vector by enforcing SmartScreen trust verification before executing Control Panel applets, but one issue went unnoticed: Explorer still automatically connects to an attacker's SMB server when displaying a folder's contents.

The Microsoft patch moved the trust verification (`ControlPanelLinkSite`, `IVerifyingTrust::OnVerifyingTrust`) only to the execution stage via `ShellExecuteExW`, but didn't fix the execution path during enumeration, tooltip display, or preview.
Because of a bug in `shell32.dll`, the `_IDCONTROLW` structure for Control Panel items (`Control Panel applet`) can trigger a call to `PathFileExistsW` by viewing a folder containing a `.lnk` file. This causes Explorer to initiate an SMB connection and leak the NTLMv2 hash before any user interaction and regardless of SmartScreen.

Article: https://nefariousplan.com/posts/cve-2026-32202-trust-check-at-the-click

PoC: https://github.com/virus-or-not/CVE-2026-32202

Akamai researchers identified CVE-2026-32202, a Windows Shell vulnerability bypassing SmartScreen. Viewing folders with malicious .lnk files triggers SMB connections, leaking NTLMv2 hashes without user interaction due to incomplete patching.

CVE-2026-32202: The Trust Check Runs at the Click. The Coercion Runs at the Icon

For informational purposes only.

Vulnerability type: Physical Access Secrets Extraction (extraction of secrets through physical access without opening the device casing)

The seller is offering an exploit for a vulnerability that allows an attacker with physical access to a VEGA3000 terminal to extract secrets stored on the device without opening the casing. The practical risk of this vulnerability is the possibility of extracting secrets (potentially encryption keys for PIN blocks, TLS certificates for the terminal, DUKPT master keys) if a cashier, service technician, or attacker with temporary access to the device gains physical access to it at the point of sale.

[VEGA3000](https://help.trustpayments.com/hc/en-us/articles/4403776573841-Castles-VEGA-3000-Pin-Pad) is a line of PIN-pad terminals from the Taiwanese company Castles Technology, running on Linux with an open architecture, certified according to PCI PTS 4.x, and supporting EMV contact/contactless. These devices are widely used in both mobile (Bluetooth/Wi-Fi/4G) and stationary configurations, and are used by merchants and integrated with payment platforms such as USAePay and Trust Payments.

Castles Technology is a Taiwanese manufacturer founded in 1993, and is one of the global leaders in the POS terminal market. According to industry reports, the top 5 global manufacturers hold over 65% of the market, and Castles Technology consistently ranks among them, alongside Ingenico, Verifone, PAX Technology, and Newland. The company claims to have [tens of millions of terminals deployed](https://www.cbinsights.com/company/castles-technology-international-corp) worldwide and partners with banks, acquirers, ISVs, and PSP providers in the retail, hospitality, gas station, and transportation sectors.

An exploit extracts secrets from Castles Technology VEGA3000 PIN pads via physical access without opening the casing, risking encryption keys if attackers gain temporary access at POS terminals.

Sale of a 0-day exploit for Castles VEGA3000 POS terminals

For informational purposes only.

Access Type: helpdesk
Revenue: $11.2 billion
Privileges: Full access to the dashboard, creation and deletion of tickets, complete details for each ticket, custom settings.
Price: $400

The seller is offering administrative access to the internal helpdesk system that handles orders for [Burger King](https://www.bk.com/). According to the seller, this access allows for the creation, preparation, and retrieval of an unlimited number of orders, provides full information on each order and ticket, and enables interaction with customers on behalf of a support representative.

Sellers offer administrative access to Burger King helpdesk for $400. This enables unlimited order creation, full ticket visibility, and customer impersonation by support representatives within the internal ordering system.

Sale of Administrative Access to the Burger King Helpdesk Panel

Strix is a multi-agent framework for automated vulnerability discovery. It is built on open-source projects such as [LiteLLM](https://github.com/BerriAI/litellm), [Caido](https://github.com/caido/caido), [Nuclei](https://github.com/projectdiscovery/nuclei), [Playwright](https://github.com/microsoft/playwright), and [Textual](https://github.com/Textualize/textual). Strix runs in a Docker environment and requires an LLM API key, while also supporting local models.

Features:
• Full penetration testing suite: HTTP proxy, OWASP Top 10 checks, SAST/DAST, OSINT, and PoC generation in a sandbox
• Multi-agent orchestration: parallel testing and knowledge sharing between agents
• Real-world validation: each vulnerability includes a working PoC and reproduction steps
• Integrations: GitHub, GitLab, Bitbucket, Slack, Jira, and CI/CD
• Automatic fix and report generation

Tool: https://github.com/usestrix/strix

Strix is a multi-agent framework for automated vulnerability discovery featuring penetration testing, orchestration, PoC generation, and integrations with CI/CD pipelines and issue trackers within Docker environments.

Strix — an open-source AI agent–based tool for penetration testing and DevSecOps

[Damn Vulnerable Drone](https://github.com/nicholasaleks/Damn-Vulnerable-Drone#installation) is a virtually simulated environment designed for offensive security professionals to safely learn and practice drone hacking techniques. It simulates real-world ArduPilot & MAVLink drone architectures and vulnerabilities, offering a hands-on experience in exploiting drone systems.

The platform operates on the principle of Software-in-the-Loop (SITL) with the Gazebo 3D robotics simulator, providing a realistic environment and physics engine for the drone.

Key features:
• runs in a docker-based setup, making it accessible and safe for drone hacking experimentation;
• simulated Wi-Fi (802.11) interfaces to practice wireless drone attacks;
• simulated RTSP drone onboard camera stream with gimbal and companion computer integration;
• companion Computer configuration management via web interface and simulated serial connection to Flight Controller;
• QGroundControl UI launching with MAVProxy GCS integration;
• dynamic Ardupilot flight bin logs stored on a simulated SD Card;
• management Web Console used to trigger scenarios and drone flight states.

The platform also includes comprehensive attack scenarios from reconnaissance (e.g., Wi‑Fi analysis, drone discovery, Companion Computer scanning) to firmware attacks (e.g., Firmware Modding, Decompile). Detailed walkthroughs for each scenario are available in the [project's wiki](https://github.com/nicholasaleks/Damn-Vulnerable-Drone/wiki/Attack-Scenarioshttps://github.com/nicholasaleks/Damn-Vulnerable-Drone/wiki/Attack-Scenarios).

Damn Vulnerable Drone is a simulated environment for security professionals to safely practice hacking drones. It uses Docker, Gazebo, and ArduPilot to simulate vulnerabilities and attack scenarios.

Damn Vulnerable Drone is a project for learning drone hacking

Researchers from Cobalt [analyzed](https://www.cobalt.io/hubfs/PDF/Cobalt-AI-Pentesting-Pulse-Report-2026.pdf?utm_campaign=45757287-2026%2006%20AI%20Pentesting%20Pulse%20Report&utm_medium=email&_hsenc=p2ANqtz-_K-_qc-2vUfx-YfzHW2YtmsC6I8eFVm0Yhm3f_9n5jkhJBCAXWZN2SwAzcbrD7ftkT_G2yQiMSeaLrGAeHI-T9_hcYfiT3vFKWyKRGa9Pswaefw6s&_hsmi=423431894&utm_content=423431894&utm_source=hs_automation) the results of five years of pentesting and two security professional surveys to assess the current state of application security in systems using AI technologies. The key takeaway from the Cobalt AI and Pentesting Pulse Report 2026: AI adoption is outpacing the development of its protection practices.

32% of findings in AI/LLM application assesments fall into the high-risk category. By comparison, across all other tested asset types this figure is around 12%. Researchers note that integrating AI tools into applications introduces new, AI-specific risks layered on top of traditional vulnerabilities.

Fixing vulnerabilities related to AI tools remains a serious challenge. The remediation rate has grown from 21.1% to 38.4%, but AI/LLM systems still show the lowest patch rate of all tested categories (for example, web apps show 73.7%).

At the same time, researchers observed a decline in trust toward automated pentests. While in 2025, 29% of organizations were ready to rely fully on automated tools, by 2026 only 9% remained. The reason is negative experience: 78% of companies encountered cases where automated scanners missed critical vulnerabilities.

More detailed survey results reflecting business responses are available in the full [report](https://www.cobalt.io/hubfs/PDF/Cobalt-AI-Pentesting-Pulse-Report-2026.pdf?utm_campaign=45757287-2026%2006%20AI%20Pentesting%20Pulse%20Report&utm_medium=email&_hsenc=p2ANqtz-_K-_qc-2vUfx-YfzHW2YtmsC6I8eFVm0Yhm3f_9n5jkhJBCAXWZN2SwAzcbrD7ftkT_G2yQiMSeaLrGAeHI-T9_hcYfiT3vFKWyKRGa9Pswaefw6s&_hsmi=423431894&utm_content=423431894&utm_source=hs_automation).

_The report stresses that as the remediation window for traditional vulnerabilities continues to shrink, the emergence of a new class of AI-specific flaws adds extra strain to security processes. Dependence on external AI providers and the lack of specialized expertise to fix AI-related issues make protecting such systems significantly harder. Therefore, when integrating AI tools into existing applications, organizations must strike a balance between leveraging new capabilities and maintaining effective risk management._

Cobalt's 2026 report indicates AI adoption outpaces security. Thirty-two percent of findings are high-risk, remediation is low, and trust in automated pentesting dropped due to missed vulnerabilities.

AI/LLM applications are becoming one of the riskiest assets

T3MP3ST orchestrates reconnaissance, exploitation, and reporting stages via CLI or a War Room browser, turning the connected AI agent (Claude Code, Codex, Hermes) or local LLM into a guided vulnerability‑discovery tool.

Capabilities:
• Multi‑agent architecture: 8 operators aligned with MITRE ATT&CK and Cyber Kill Chain phases.
• Integration with local OpenAI‑compatible models without API keys (Ollama, LM Studio, vLLM).
• Automatic result verification using `npm run verify-claims` and recalculation of all metrics based on execution artifacts.
• Built‑in safeguards against out‑of‑scope actions.
• API, MCP server, and an extensible arsenal of 35 tools (83 in the full set).

Compared to similar frameworks (e.g., `Metasploit Framework`, `Cobalt Strike`) `T3MP3ST` stands out with its cloud‑free, API‑key‑independent architecture and native LLM agent integration. However, the platform remains experimental: the effectiveness of its multi‑agent coordination has yet to be validated by comprehensive benchmarks, unlike traditional C2 frameworks.

Tool: [https://github.com/elder-plinius/T3MP3ST](https://github.com/elder-plinius/T3MP3ST)

T3MP3ST is an experimental AI framework orchestrating vulnerability discovery via local LLMs and multiple agents. It offers offline operation distinct from traditional C2 tools but lacks validated benchmarks.

T3MP3ST — framework for automated security analysis powered by an AI coding agent

Major update adds new features, improves the architecture, and fixes several bugs.

New features:
• Download SYSVOL share contents with filtering capabilities.
• Export LDAP data directly into an SQLite database.
• Parse individual GPO files.

Tool: [https://github.com/cogiceo/GPOHound](https://github.com/cogiceo/GPOHound)

The GPOHound tool received major update with architecture improvements and bug fixes. New capabilities allow SYSVOL filtering, LDAP export to SQLite, and parsing individual GPO files.

New version of the GPOHound

Researchers from the Czech Republic [analyzed](https://www.helpnetsecurity.com/2026/07/03/research-non-interactive-ssh-attacks/) activity from 11 SSH honeypots and examined attacker behavior after a successful login. Over 15 days, they recorded 117,622 SSH sessions, and 99.23% of them were non‑interactive.

This means that after authentication, the client sends one or two commands via an SSH exec request, the server executes them without launching an interactive shell, and then instantly closes the connection once the result is returned.

Only in 0.1% of cases was an interactive shell launched — the kind of session where attacker behavior can actually be observed. Traditionally, SSH honeypots have focused on such sessions. The rest of the activity comes from automated scripts and scanners:

• The 10 most frequent commands account for 41.59% of all traffic and are primarily used to gather basic system information.
• 2,178 sessions showed bots checking whether the commands were really executed on the server (i.e., testing whether the environment is "real" or not).
• Some requests tried to detect the presence of a honeypot itself — for example, by looking for Cowrie or Kippo processes or checking system file availability.

Historical data from CZ.NIC (the Czech ccTLD registry) indicates this isn't a new phenomenon: an observation archive since 2017 (over 400 million sessions) shows that non‑interactive traffic has dominated since 2018.

_The main takeaway from the research is the challenge of measuring honeypot effectiveness. It's often gauged by how long an attacker stays in the system or how many commands they execute. In reality, nearly all traffic consists of single requests completed in fractions of a second, making such metrics unreliable for assessing true threat activity. Therefore, SSH honeypot logic should shift toward detecting patterns in these short interactions to identify broader attack campaigns._

Czech study analyzed 117,622 SSH sessions; 99.23% were non-interactive. Most traffic comes from automated scripts. Traditional honeypot effectiveness metrics are unreliable.

SSH honeypots mostly capture bot activity

In the article, the researcher demonstrates how a use-after-free bug in the `DRM_IOCTL_GEM_CHANGE_HANDLE` handler in the Linux kernel lets a regular user gain root privileges ([CVE-2026-46215](https://t.me/c/3422352556/2/1525)). When a GEM object is transferred between handles, the `handle_count` counter isn't updated, and if `DRM_IOCTL_GEM_CLOSE` is invoked simultaneously, the old handle frees the object, leaving the new one pointing to freed memory.

The exploit leverages a race condition to overwrite the freed portion of the `drm_gem_object` structure with `pipe_buffer` elements (via a pipe/msg_msg spray), leaks a kernel pointer to bypass KASLR, and sets `PIPE_BUF_FLAG_CAN_MERGE` to once again bypass the DirtyPipe fix. As a result, an attacker with access to `/dev/dri/renderD` can modify `/etc/passwd` from a user process and gain root access.

Article: https://cyberstan.co.uk/drm-lpe-linux/

PoC: https://github.com/0xCyberstan/CVE-2026-46215-POC

Use-after-free bug in DRM_IOCTL_GEM_CHANGE_HANDLE in Linux kernel allows root privileges via race condition.

Analysis of vulnerability CVE-2026-46215: local privilege escalation via UAF in Linux DRM GEM

The research demonstrates an interrupt descriptor table (IDT) substitution technique in Windows 11 that works even with VBS/HVCI protections enabled. The method follows a "data‑only" approach: the original physical IDT page is cloned into a free writable area (`FWA`), where the required descriptor is modified, and then the IDT page mapping is briefly switched to this clone. After the payload (`token‑swap payload`) completes, the mapping is restored to its original configuration.

The manipulation leverages the `INT 2E` vector as a bridge to an internal kernel service function, temporarily redirecting the `NtSetQuotaInformationFile` slot to the target handler. This enables calls to `PsGetCurrentProcess`, `PsReferencePrimaryToken`, `memcpy`, `ObDereferenceObject`. This achieves kernel-level code execution without altering the original IDT or compromising memory integrity, confirming that the IDT structure can serve as a data redirection surface.

 Article: https://www.exploitpack.com/blogs/news/idt-table-hijacking-under-vbs-hvci-kcet-in-windows-11

Technique allows IDT substitution in Windows 11 with VBS/HVCI protections enabled using 'data-only' approach.