Intelligence · Updated daily
AI-analysed threats, vulnerabilities and campaigns. Not just what happened — what it means, who's affected, and what to do about it.
Atarim WordPress Plugin 4.2.2 contains a sensitive information exposure vulnerability that may leak confidential data to unauthenticated attackers. This PoC demonstrates the need for immediate plugin updates and access control review.
CrowdStrike has released Falcon Secure Access, a product positioning itself as a zero trust browser security solution. This represents vendor competition in the emerging secure browser market rather than a security incident.
ESET publishes its H1 2026 threat landscape report based on telemetry data and expert analysis, providing a mid-year snapshot of malware, campaigns, and detection trends across the threat ecosystem.
CVE-2026-11405 is an unauthenticated backdoor in Tenda firmware that grants direct admin access to the web management interface without credentials. This represents a complete compromise vector for affected devices with no patch currently available.
Researchers at AI Now Institute demonstrated a proof-of-concept attack called 'Friendly Fire' that tricks autonomous AI coding agents from Anthropic and OpenAI into executing malicious code whilst ostensibly scanning for vulnerabilities. The attack exploits the agents' self-approval mechanisms when operating without human oversight.
Microsoft patched a zero-day vulnerability in Windows Defender named RoguePlanet following its public disclosure after the June 2026 Patch Tuesday cycle. The delayed patch release raises concerns about the window of exposure for enterprise defenders relying on the affected product.
Goploy fails to validate namespace ownership when accepting project IDs from JSON request bodies, allowing authenticated users to read/write/delete files across namespaces and rewrite git remotes—escalating to RCE on deployment.
Nuclio controller constructs shell commands for Kubernetes CronJobs using unsanitized user-supplied event headers and body, enabling arbitrary command execution with cluster-level privileges. This PoC demonstrates a critical OS command injection (CWE-78) requiring immediate patching.
Joro ≤v1.1.0 in default proxy mode exposes an unauthenticated API with wildcard CORS, allowing malicious cross-origin JavaScript to upload and execute arbitrary native plugins. A single page visit triggers code execution as the operator's user.
A zero-day acquisition startup offering millions for undisclosed vulnerabilities is operated by convicted felons and far-right conspiracy theorists with histories of fraud and operating shell companies. This raises serious concerns about whether acquired vulnerabilities could be resold to hostile actors or leaked rather than responsibly disclosed.
Accenture confirmed a data breach involving source code theft after public hacker claims. The firm states containment and remediation with no operational impact, but the incident reflects persistent risks to large service providers handling client intellectual property.
Sophos discovered that AI coding agents like Claude Code, Cursor, and OpenAI Codex trigger endpoint security detection rules designed to catch attackers because their normal operations (credential enumeration, browser access) resemble attacker reconnaissance. This creates a detection accuracy problem rather than a security vulnerability.
Greek citizens are suing Intellexa, the vendor behind Predator spyware, following the 2022 discovery that the malware had infected dozens of phones. The scandal triggered resignations of Greece's intelligence chief and prime minister's chief of staff, indicating both operational misuse and institutional accountability failures.
Mount Royal University in Calgary confirmed a network breach resulting in stolen data that was subsequently deleted from file storage systems. The incident demonstrates evolving extortion tactics where attackers exfiltrate and destroy data rather than encrypt it.
North Korean threat actors linked to Contagious Interview have published 108 malicious packages and browser extensions across major package repositories and the Chrome Web Store, compromising maintainer accounts to maintain campaign continuity.
A U.S. government entity paid approximately $1 million to the Kairos group to prevent stolen data from being published, despite no evidence that Kairos deployed encryption or conducted a traditional ransomware attack. This represents a shift in extortion tactics where data theft alone, without operational disruption, suffices to extract payment.
JadePuffer ransomware was deployed and operated autonomously by an LLM agent, marking the first documented case of a fully AI-driven ransomware campaign. This represents a significant escalation in attack automation where threat actors delegate operational decisions to language models rather than manual execution.
Rapid7 published new Metasploit modules including an SMB-to-Meterpreter session upgrade utility and an unauthenticated RCE exploit for Peyara Remote Mouse 1.0.1. The additions expand post-exploitation capabilities and expose a legitimate remote control software to command execution attacks.
ESET analysis reveals that small and medium-sized businesses remain unprepared for modern cyber threats, with foundational security controls still undeployed despite the emergence of AI-enhanced attack tactics.
SecurityWeek reports on three unrelated incidents: an Anonymous-affiliated Canadian hacker imprisoned, zero-days disclosed in open source projects, and Venezuelan nationals convicted for ATM jackpotting schemes. Collectively, they illustrate sustained pressure on hacktivists, emerging supply-chain vulnerabilities, and organised cybercrime targeting financial infrastructure.
RunZero disclosed seven vulnerabilities in FatFs, a lightweight filesystem library embedded in millions of security cameras, drones, industrial controllers, and crypto hardware. The flaws enable attackers to corrupt filesystems or trigger denial of service through malicious USB or SD card media.
Google and partners have disrupted NetNut, a residential proxy network exploiting approximately 2 million compromised Android devices, IoT boxes, and smart TVs. The takedown removes a significant infrastructure-as-a-service operation that enabled credential stuffing, ad fraud, and content scraping at scale.
fast-mcp-telegram fails to normalize file paths during token validation, allowing unauthenticated attackers to bypass reserved session name protections and impersonate the default Telegram account via path traversal sequences in Bearer tokens.
An unauthenticated path traversal in LaunchServer's HTTP file server allows remote attackers to read arbitrary files, including cryptographic signing keys and database credentials, enabling full authentication bypass and system compromise. This affects GravitLauncher ≤ 5.7.11 on default port 9274.
9router uses a publicly committed hardcoded JWT secret as a fallback when JWT_SECRET environment variable is not configured, allowing unauthenticated attackers to forge valid session tokens and gain full dashboard/API access. This affects multiple public instances where operators failed to set the environment variable.