CWE-bench v2 raises the bar for AI security evaluation
Collinear AI releases 100 open-source tasks to test how well models find and fix real code vulnerabilities without breaking functionality.
Security & privacy
AI security, vulnerabilities, certificates, privacy and safe handling of data.
Google dismantles XCPE botnet infecting 1.3 million Android TV boxes
Google and partners took down the XCPE botnet, which turned over 1.3 million Android TV boxes into residential proxies for ad fraud and traffic anonymization.
Critical Linux TCP flaw allows remote code execution without authentication
Red Hat and Ubuntu address CVE-2026-98432, a critical use-after-free vulnerability in the Linux kernel IPv4 TCP stack that enables remote code execution.
AI-generated scripts and SysGoRAT target government Linux servers
Palo Alto Networks Unit 42 details the CL-UNK-1211 campaign, where attackers used AI tools to generate malicious scripts and deploy SysGoRAT on government Linux infrastructure.
Volta malware uses AI to rank stolen data by financial value
Cisco Talos researchers identified VOLTA-AUTODOXER, a Windows infostealer that leverages LLMs to analyze and prioritize victim data for monetization.
Critical Langflow vulnerabilities allow unauthenticated remote code execution
IBM reports two critical flaws in Langflow versions 1.0.0 through 1.12.8 that enable authentication bypass and arbitrary code execution. Version 1.13.0 is the only fixed release.
North Korea's TraderTraitor linked to record $6.43 billion Binance heist
The FBI attributes the largest crypto theft in history to a North Korean group using trojanized Terraform providers and AI-driven social engineering.
Polymarket loses $2.7m after DebugBear script hijack via DNS compromise
Hackers stole funds from Polymarket traders by injecting malicious code through a compromised DebugBear monitoring script, leading to a $3.9 million refund effort.
NetamLoader malware uses .NET injection to target Asian government Windows systems
Kaspersky researchers uncovered NetamLoader, a sophisticated .NET malware framework targeting custom Windows 10 government builds in Asia via supply chain compromise.
Critical LXD and MicroCloud flaw allows root access without private keys
A CVSS 10.0 vulnerability in Canonical's LXD and MicroCloud lets attackers impersonate trusted clients using only public certificates, potentially granting root access to host machines.
ChatGPT share links allowed code execution in user sessions
Researchers found a vulnerability in ChatGPT share links that let attackers run code and read private chats. OpenAI fixed the flaw within eight days.
Ivanti Connect Secure flaws chained for unauthenticated root access
CISA adds two Ivanti Connect Secure vulnerabilities to its KEV catalog after active exploitation by threat actor UTA0576 using a three-flaw chain.
Critical Linux TCP vulnerability CVE-2026-98432 requires immediate patching
A critical use-after-free bug in the Linux kernel's IPv4 TCP handshake allows remote code execution. Upstream fixes are available, but major distributions like Ubuntu and RHEL have not yet released pa
How a shared ChatGPT link could hijack user accounts via sandbox escape
Researchers found a chain of vulnerabilities in ChatGPT's HTML Preview and MathLive integration that allowed attackers to execute code in victim sessions via public share links.
TeamViewer flaw lets controlled hosts attack viewer machines
A high-severity memory overflow in TeamViewer allows a remote host to crash or execute code on the controlling viewer. Update to version 15.83 immediately to patch CVE-2026-111585.
