Researchers from Anthropic and EPFL have demonstrated that self-propagating payloads can spread between autonomous AI agents through editable system prompt files used for state persistence. The technique, tested in a simulated six-agent coding environment, raises concerns about the integrity of autonomous AI systems.
Cybersecurity researchers at Ontinue have disclosed a previously undocumented Python implant framework called TWINLOOT. The malware is hardened with PyArmor and operates its entire command-and-control (C2) infrastructure through SharePoint Online and Microsoft Teams, making detection difficult by blending in with legitimate traffic. The framework is designed for credential theft and lateral movement across networks.
Research published by the agent security platform Reco identifies a sustained scraping campaign, dubbed 'City Forum,' targeting Salesforce and ServiceNow customer portals across multiple industries. The activity is traced to a single IP address, 158.220.87.79, with reports indicating the operation has been active since at least 2025. The exact method of access and specific victim data remain under investigation.
A typosquatting campaign has been identified distributing malicious RubyGems packages designed to steal browser-stored credentials and cryptocurrency wallet data from Windows users. Researchers from OpenSourceMalware are tracking the threat under the moniker "StubMaker." The campaign includes packages with names intentionally similar to legitimate projects, relying on user error during installation. Analysis indicates the packages execute an information stealer upon installation, targeting data stored in browsers and cryptocurrency wallet applications. The discovery was reported on August 18, 2026.
SafePal has disclosed that an authorization flaw in an order-tracking plug-in exposed the names, email addresses, shipping addresses, phone numbers, and purchase details of approximately 39,798 customers. Affected individuals were notified via email on August 16, 2026. The company states no evidence of active exploitation has been reported, and the plug-in authorization flaw has been patched.
The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has added a critical vulnerability impacting the Ray distributed computing framework to its Known Exploited Vulnerabilities (KEV) catalog. The inclusion is based on evidence of active exploitation in the wild. The framework, which is Python-native and open-source, is widely used to scale artificial intelligence and machine learning workloads. While the advisory confirms active exploitation, specific technical details such as the CVE identifier, precise attack vector, and scope of impact remain limited in the initial reporting.
Researchers from Wiz have disclosed a GitHub Actions workflow injection vulnerability in Snowflake's public snowflakedb/snowflake-connector-net repository. The flaw, located in the .github/workflows/jira_issue.yml workflow, could potentially allow command execution via crafted GitHub issues. The workflow was designed to run on issue activity and contained internal Jira credentials, raising concerns about credential exposure and lateral movement if exploited.
Kaspersky researchers have identified new components in the Cavern (Cav3rn) command-and-control framework used by Iranian nation-state hackers. The framework has evolved to leverage DNS tunneling and Google Apps Script to blend into legitimate traffic, targeting entities in Israel. The ongoing monitoring since December 2025 has revealed previously unreported capabilities that expand the threat actor's operational reach.
During the week of August 17, 2026, cybersecurity incidents included exploitation of VMware ESXi vulnerabilities, active exploitation of a Windows zero-day, attacks targeting the Model Context Protocol (MCP), and browser hijacking campaigns. The reported events highlight that high-cost attacks are not always the most sophisticated, with exposed services, legacy bugs, and supply-chain trust issues serving as primary entry points.
A new analysis highlights how MCP servers can expose enterprise secrets through plaintext configuration files, over-permissioned access, and prompt injection. As organizations increasingly adopt AI agents, these exposures can silently become a major security gap, often before security teams are aware the server is running.
Security researchers from SSD Secure Disclosure have detailed a two-stage exploit chain that achieves full Android kernel access on devices using Unisoc modem firmware through a VoLTE video call. The advisory, published August 17, 2026, represents the second stage of a chain first disclosed in March 2026, which revealed remote code execution in the Unisoc modem firmware. The second stage escalates privileges to kernel level. Unisoc has not provided a fix as of publication.
A recent advisory from ThreatLocker, reported by BleepingComputer, highlights that Windows named pipes—used for interprocess communication—can expose privileged services to untrusted processes when access controls are weak. The guidance outlines defensive measures including endpoint verification, command authorization, strict input validation, and narrowly scoped privileges to reduce exploitation risk.
A supply-chain attack targeting Android-based car head units is leveraging a legitimate device-update app to distribute malware that turns compromised devices into a proxy botnet or uses them for ad fraud. The attack was reported on August 22, 2026, and affects users globally, though specific regions are not disclosed. The malware is delivered during the update process, potentially degrading device performance, consuming data, and exposing users to further malicious activity. No specific patch is available; users and manufacturers are advised to audit update channels and monitor for unusual network behavior.
Researchers have identified Evooo1Bot, a Linux botnet family derived from the publicly leaked Mirai source code, which exploits known vulnerabilities in edge‑network devices to conscript them as SOCKS5 proxies. The malware retains the original DDoS functionality while extending the framework with proxy‑routing capabilities. No zero‑day vulnerabilities have been reported; the attacks rely on unpatched known flaws.
The Netherlands National Cyber Security Centre (NCSC-NL) has issued an advisory warning that a critical authentication flaw in Apple macOS Screen Sharing (CVE-2026-65400, CVSS 9.8) is being actively exploited in the wild. Attackers are leveraging the vulnerability to gain unauthorized access to internet-exposed macOS systems and deploy a Monero cryptocurrency miner. Apple has released security updates to address the vulnerability, and users are strongly advised to apply the latest macOS updates immediately.
Threat actors spent approximately $7 million during the first half of 2026 to acquire expired domains through dropcatching techniques. By inheriting existing web traffic, SEO rankings, and backlink profiles, these domains are being used to redirect victims to scams and malware on a large scale. DNS threat intelligence firm Infoblox identified 50,400 such domains registered in the period, highlighting a growing abuse of the domain lifecycle for malicious infrastructure.
Kaspersky reports that the threat actor known as Mustang Panda (aka HoneyMyte) has updated its CoolClient backdoor with a signed Windows kernel-mode rootkit. The rootkit is capable of hiding malicious processes, files, registry objects, and C2 network information from detection. Victims have been identified in Myanmar, Mongolia, and Pakistan, though the full scope and infection vectors remain under investigation.
Researchers have detailed a post-exploitation technique that enables the Chrome DevTools Protocol (CDP) inside a running Google Chrome or Microsoft Edge process on Windows. The method allows an operator with existing code execution on the host to access cookies, saved data, and authenticated browser sessions. The technique does not require user interaction and leverages the DevTools protocol to interact with the browser process.
Researchers from CTM360 have identified a large-scale, global phishing campaign themed around recruitment. The operation uses fake interview scheduling pages and Browser-in-the-Browser (BitB) windows to harvest Google and Facebook credentials. In more sophisticated instances, the campaign relays multi-factor authentication (MFA) prompts in real time to facilitate account compromise. The activity was detailed in a report titled RecruitTrap, published by CTM360 and covered by The Hacker News on August 14, 2026.
Apple has sent new threat notifications to users in 110 countries, alerting them that they may be targeted by mercenary spyware attacks. The company stated it has now notified customers in over 150 countries to date, with notifications beginning in late 2021. Apple did not attribute the attacks to specific threat actors or provide technical details about the exploit vectors used.
The China-linked threat actor known as Jewelbug has been observed conducting cyber espionage operations targeting governments and militaries while simultaneously engaging in cryptocurrency fraud. Both malicious campaigns are administered from a single control panel, XG-Web, a browser-centric remote-access and information-stealing framework that transforms a victim's browser into a full remote-control tool. The operations were reported by The Hacker News on August 14, 2026.
The latest ThreatsDay Bulletin from The Hacker News provides a comprehensive overview of this week's diverse cybersecurity landscape, featuring GhostJacking AI attacks that exploit AI system resources, EtherHiding ClickFix campaigns leveraging social engineering through fraudulent click-fix guides, and a Cursor CLI flaw posing potential command execution risks. The bulletin also includes 17 additional updates spanning cloud service misconfigurations, malware variants, data breach disclosures, and various scam operations targeting global users and organizations.
Acronis Threat Research Unit has identified an ongoing cyberattack campaign delivering a new backdoor named PATCHCORD. The implant, written in C/C++, targets Afghan telecom providers and South Asian critical infrastructure organizations through sector-specific lures, including fake VPN installers impersonating Afghan Telecom. The campaign aims to establish unauthorized access for potential espionage and data exfiltration.
Researchers have identified AmnesiaStealer, a new macOS-oriented information stealer written in Rust, capable of hijacking Chromium-based browsers to steal session data and potentially grant live browser control. The malware is distributed through fraudulent GitHub download pages that impersonate legitimate software, employing ClickFix-style social engineering to trick users into executing the payload.
