A six-month investigation uncovered over 30 vulnerabilities across every major AI-integrated IDE tested, including GitHub Copilot, Cursor, Claude Code, and JetBrains products, resulting in at least 24 assigned CVEs. The attack chain exploits prompt injection via rule files, READMEs, or malicious MCP servers to hijack AI agents, which then abuse legitimate IDE features like JSON schema fetching or executable path configuration to exfiltrate data or achieve RCE. Organizations using AI coding assistants should audit workspace configurations, restrict AI agent permissions, and monitor for suspicious outbound requests from development environments.
Security researcher Jeremiah Fowler uncovered an unsecured, publicly accessible database containing images and videos that appeared to belong to a company named SocialBook. The database included pornographic face swaps, deepfakes, and unaltered reference photos. It is unclear how long the images were accessible or whether the exposure was caused directly by SocialBook or through a third party.
Petco reports that a misconfigured software application briefly left certain customer files accessible online, exposing unspecified personal data before the issue was fixed. Details on the number of victims and exact data types remain undisclosed, though legal thresholds suggest that at least hundreds of customers were impacted.
This framework outlines six principles for browser-centric zero trust architecture: identity-first access control using federated IdPs, least-privileged access with time-bounded entitlements, continuous verification with adaptive policies, phishing-resistant FIDO2/WebAuthn authentication, device health gating via MDM/EDR integration, and remote browser isolation for high-risk sessions. Implementation follows CISA's Zero Trust Maturity Model through three stages—starting with SSO and passkey MFA, advancing to device compliance enforcement with SCIM automation, and achieving optimal state with per-request access decisions and automated RBI for privileged operations. Security teams should prioritize deploying passkeys to administrators first, integrate IdP conditional access with SIEM/SOAR to automate session revocation, and codify all ZTNA policies as infrastructure-as-code for governance.
Unit 42 researchers demonstrated three attack vectors exploiting the Model Context Protocol (MCP) sampling feature in AI coding copilots: resource theft through hidden prompts that consume token quotas invisibly, conversation hijacking via persistent prompt injection that alters LLM behavior across sessions, and covert tool invocation enabling unauthorized file system operations without user awareness. The attacks exploit MCP's implicit trust model, where malicious servers can craft prompts requesting LLM completions, with responses processed and hidden from users by the client's summarization layer. Organizations should implement request sanitization with strict templates, enforce capability declarations limiting server permissions, require explicit user approval for tool executions, and deploy statistical analysis to flag abnormal token usage patterns.
2025.12.08·
unit42.paloaltonetworks.com·15 min read·TLDR InfoSec
TangleCrypt is a previously undocumented Windows malware packer that features complex features such as the ability to execute in its own process or a child process based on a user-supplied string. The packer employs several common anti-analysis techniques such as anti-debugging, string encryption, dynamic import resolving, and unoptimized code. There are several bugs in the packer, which suggests that it may still be a work in progress.
When identity breaks down, the consequences ripple across the business. If you feel like manual IAM processes aren't keeping up with modern demands, grab this free Tines guide [11] for a 5-step IAM framework, practical checklist, and pre-built workflows for a more scalable IAM strategy [11].
EvilMist's Enumerate-EntraUsers is an Azure Entra ID reconnaissance tool featuring 15+ user enumeration methods designed to work even when direct /users API access is blocked. It assesses MFA gaps, conditional access policy weaknesses, and privilege escalation paths with BloodHound-compatible export. Available in PowerShell and Python with stealth mode (configurable delays/jitter) for detection evasion.
Helmet Security provides an end-to-end security platform for agentic AI, focused on securing Model Context Protocol (MCP) servers and connections between AI agents, software, and data. It discovers MCP endpoints, monitors and logs traffic, and blocks noncompliant or risky AI communication paths.
Researchers from Amnesty International and a coalition of newspapers released a series of reports based on data from Intellexa, a company selling spyware to governments and other clients. The leaked data included training videos showing that Intellexa employees could remotely access some of their customers through TeamViewer. Intellexa denies any misconduct, asserting that it only uses TeamViewer with client permission and supervision for technical support when issues need fixing.
Given the non-deterministic nature of LLMs, security teams need to embed themselves in the development of agentic and other AI tools. They should collaborate with developers and share tooling for code analysis and agent monitoring. Security teams should also use AI-based tools and social engineering to pentest models.
Cloudflare briefly knocked many customer sites and services offline after a flawed Web Application Firewall update meant to block a new remote code execution vulnerability in React Server Components. The misconfiguration took Cloudflare's network down for several minutes, disrupting platforms from e-commerce to cloud services and echoing earlier large outages at Cloudflare and AWS.
Banks, hospitals, and telecoms run on mainframe. Big Iron Bits: a Mainframe Mockumentary shows why this engine of the future keeps thriving. Laugh your way through mainframe fact and fiction. Start
A bipartisan group of senators reintroduced the Health Care Cybersecurity and Resiliency Act, which would mandate updates to HIPAA regulations, authorize a five-year grant program for health care entities, and improve HHS-CISA coordination on incident response and training.
Apple and Google issued threat notifications to users in over 150 countries, with Google specifically warning that several hundred accounts were targeted by US-sanctioned Intellexa spyware, exploiting zero-day vulnerabilities in mobile browsers.
China-based SMS phishing crews are sending iMessage and RCS texts about rewards points, tax refunds, and fake retailers to steal victims' card data and one-time codes.
ThreatLocker's annual Zero Trust World [4] is the most interactive, HANDS-ON CYBERSECURITY LEARNING event. Join hacking labs, get Cyber Hero certified, and attend sessions with cybersecurity, IT, and business experts.
Marquis, a Texas-based fintech that provides marketing and compliance services to over 700 US banks and credit unions, suffered a ransomware attack on August 14 that exploited a zero-day vulnerability in SonicWall. Attackers stole sensitive customer data, including names, contact details, financial account numbers, and Social Security numbers, impacting at least 400,000 people so far. The total is expected to grow as more state notifications are filed.
A critical flaw in React 19, dubbed React2Shell, allows attackers to execute code remotely on applications that use newer server-side features and common frameworks. Many cloud environments are reportedly exposed. Vendors are already shipping detection and protection rules.
Freedom Mobile, Canada's fourth-largest wireless carrier, disclosed a data breach impacting an undisclosed number of customers. The breached data includes first and last names, home addresses, dates of birth, home and/or cell phone numbers, and Freedom Mobile account numbers. The data was breached via the account of a subcontractor. Ransomware was not involved.
Cotool tested an AI agent's ability to solve security operations tasks using the Splunk BOTSv3 dataset and frontier models from Claude, OpenAI, and Gemini. Cotool found that while GPT-5.1 and Opus 4.5 achieved the top accuracy at 63%, Opus was 3x more expensive than GPT-5.1 but also completed tasks in half the wall-clock time of any other model. The Gemini models performed notably poorly on the tasks and failed to complete several tasks.
Ransomware operators often use remote access tools such as AnyDesk and UltraViewer to maintain persistence, execute remote tasks, and transfer files while avoiding detection. Their typical attack involves using stolen credentials to compromise an employee, then either hijacking existing tools or installing legitimate remote access software, followed by establishing persistence, escalating privileges, and disabling security tools. Subsequently, they deploy payloads, move laterally within the network, and finally execute their ransomware attack.
Researchers trapped North Korean operatives from Lazarus Group's Famous Chollima division inside an extended ANY.RUN sandbox environments, documenting their full operational workflow for infiltrating Western companies as remote IT workers. The operators recruit developers via GitHub spam. They offer 20-35% salary cuts in exchange for using victims' identities and laptops for job applications, relying on AI interview tools, AstrillVPN, and remote access software rather than malware. Poor operational security allowed researchers to delay, monitor, and ultimately expose the scheme targeting the finance, crypto, and healthcare sectors.
7AI builds an “agentic security” platform where autonomous AI agents investigate alerts, correlate data, and take or recommend actions to reduce alert fatigue.
GoldFactory is a cybercrime group targeting Android users in Indonesia, Thailand, and Vietnam with tampered banking apps that still appear legitimate while secretly installing remote-access malware. The malware abuses accessibility services and hooking frameworks to bypass security checks, hide its presence, control devices in real time, and steal banking data at scale. At least 11,000 infections have been identified so far.
Arizona Attorney General Kris Mayes has joined several other states in suing Temu and its parent company over allegations that the online retailer is stealing customers' data. Mayes stated that the app collects a startling amount of data without users' consent, specifically citing GPS location and a list of other apps on users' phones. Mayes also expressed concerns over the quality of products sold on Temu and whether the retailer has “ripped off” other businesses.
Anthropic evaluated ten frontier models on a dataset of 405 historical smart contract exploits and found that the agents produced working attacks against 207 of them, totaling $550M in simulated stolen funds. Anthropic measured the model's current capabilities by plotting each model's total exploit revenue against its release date. The models also discovered two previously undisclosed zero-days.
Blumira's SOC Auto-Focus translates complex findings into plain language, giving teams instant clarity, clear priorities, and guided next steps, without replacing human judgment. Learn more about Blumira
Malwarebytes' Lock and Code podcast revisits UK research revealing that air fryer companion apps requested precise location and audio recording permissions without explanation.
A website called Proxyearth, launched in October, enables anyone to retrieve Indian citizens' full names, Aadhaar numbers, residential addresses, and live locations using only a mobile number.
Today's social engineering spans far beyond the email inbox. Bad actors use AI to phish, and are using SMS texts, voicemails, and even video chats to scam.
The Everest ransomware group claimed responsibility for breaching ASUS and exfiltrating over 1TB of data allegedly containing proprietary camera source code, firmware, and drivers for devices with built-in cameras. The group issued a 21-hour deadline for ASUS to respond via Qtox encrypted messaging, following recent claimed attacks on Under Armour, Petrobras, and Iberia airline. ASUS has neither confirmed nor denied the breach, which, if verified, would mark the second significant infrastructure compromise since the 2019 ShadowHammer supply chain attack.
Financial authorities have begun an emergency on-site inspection of the Korean e-commerce platform Gmarket after reports of unauthorized mobile payments. The attackers purchased mobile gift certificates using credit cards registered with Gmarket's simple payment service. Gmarket claims that the unauthorized payments resulted from account compromise, not a data breach.
A security researcher discovered a privilege escalation vulnerability in the K7 Ultimate Security antivirus while investigating an unrelated DoS vulnerability. The vulnerability exploited named pipes with overly permissive access control lists, allowing a non-admin user to execute arbitrary code as SYSTEM during fake updates. K7 worked with the researcher to issue three patches to provide interim fixes for the vulnerability pending full ACL enforcement in a future release.
A Windows privilege escalation technique exploits named pipes and the Print Spooler's RPC interface to elevate accounts with SeImpersonatePrivilege, such as IIS app pool identities, to SYSTEM. It exploits its lack of SQOS protection and a path-validation bug that can be bypassed by crafting backslash/forward-slash-separated pipe paths to trigger SYSTEM connections to an attacker‑controlled pipe.
Modern phishing operations increasingly rely on mainstream infrastructure like Cloudflare, major cloud providers, and popular developer-focused TLDs such as .dev and .app to appear legitimate and resist takedown. Attackers mix disposable direct-hosted servers with CDN-protected setups, often hiding behind services like Vercel, GitHub Pages, and other trusted platforms that organizations cannot simply block. Phishing-as-a-Service ecosystems now provide turnkey kits with MFA bypass, session cookie theft, and advanced evasion, heavily targeting brands like Meta, Amazon, and PayPal.
ZeroCrumb bypasses the Chrome Elevation Service by impersonating a Chrome instance to dump app bound protected credentials and cookies without privileges.
Maryland-based Frenetik emerged from stealth with patented "Deception In-Use" technology (US Patent 12,463,981) that continuously rotates active identities and resources across Microsoft Entra, AWS, Google Cloud, and on-premises environments to invalidate attacker reconnaissance. The approach exploits information asymmetry by ensuring defenders receive real-time change notifications through out-of-band channels while attackers operate on stale intelligence, funneling them into honeypots and decoys. A free community version is available for organizations seeking to implement active deception without extensive tuning or analyst oversight.
Researchers from BCA LTD, NorthScan, and ANY.RUN captured Lazarus Group's Famous Chollima division operating live by deploying sandbox VMs disguised as developer workstations to trap operators targeting Western companies in finance, crypto, healthcare, and engineering sectors. The investigation revealed a toolkit focused on identity takeover rather than malware. Organizations should implement enhanced vetting for remote hires as attackers request full identity documents and 24/7 laptop access to infiltrate companies without deploying traditional malicious payloads.
The new Apple Security Bounty payouts devalue macOS. Full Transparency, Consent, and Control (TCC) bypasses are down from $30.5k to just $5k, while other TCC categories have been reduced from $5-$10k to just $1k. This shows an admission by Apple that it can't fix everything and is prioritizing iPhone users over Mac users.
A long‑running online gambling operation targeting Indonesian users is built on hundreds of thousands of owned and hijacked domains, compromised WordPress/PHP sites, and malicious Android apps. Its cost, scale, and advanced tradecraft point to an APT-style, possibly state-aligned group that uses the gambling infrastructure as both a revenue stream and cover for stealthy access to Western government and industry networks.
Codacy's new AI Risk Hub enforces policies against GenAI-related code risks. Catch unapproved model calls, invisible unicode injections, secrets and insecure dependencies across the codebase, instantly.