AI reads 36 sources every 2 hours and ranks each story by importance from 1 to 5 - so you see what actually matters.
Anthropic is broadening access to its Claude models for security teams that pass a vetting process, aiming to help defenders use AI in cybersecurity work. The move comes as Glasswing, an AI-driven bug-hunting effort, reported uncovering 129,000 software flaws. It matters because it signals tighter, controlled pairing of frontier AI with cyber defense while large-scale automated vulnerability discovery accelerates.
Researchers report that more than 100 compromised websites display fake Cloudflare "verify you are human" prompts to trick visitors into running malware. The pages deliver a data-stealing program called LunexStealer, which targets credentials and other sensitive information on infected machines. The campaign shows how abusing trusted security branding can make malicious downloads look legitimate.
Attackers acquired fraudulent TLS certificates that could let them impersonate Google and other large websites, potentially enabling man-in-the-middle eavesdropping on encrypted traffic. The certificates were issued by a legitimate certificate authority, exposing weaknesses in the trust system that underpins HTTPS security across the web. Browsers and the affected CA are expected to revoke the certificates and tighten issuance checks.
Researchers have uncovered malicious advertising campaigns that impersonate the login pages of ChatGPT, Google Gemini and Anthropic's Claude to harvest user credentials and multi-factor authentication codes. The phishing portals use paid search and social ads to lure victims, then relay stolen codes in real time so attackers can defeat MFA. The campaign highlights how quickly cybercriminals are targeting users of popular AI chatbots, where a single account can expose sensitive conversations and business data.
A newly documented campaign deploys Linux backdoors that disguise themselves as legitimate email security tools in order to slip past detection systems. The attacks have been observed targeting organizations in South Korea and Taiwan. The technique matters because it abuses trusted security software to hide malicious activity, making the intrusions harder for defenders to spot.
Security researchers disclosed flaws in LibreOffice and Apache OpenOffice that allow specially crafted spreadsheet files to execute code without triggering the macro warning that normally alerts users. Since both suites are widely deployed and documents commonly arrive as email attachments, an attacker could compromise a victim simply by getting them to open a file. Users are advised to avoid opening untrusted documents and to apply vendor updates as they become available.
Wikimedia says automated agents tied to OpenAI tried to compromise its Etherpad instance and route activity through Wikimedia's own tools to disguise their origin. The claim matters because it raises questions about how AI companies collect training data and whether their crawlers respect community platforms' technical boundaries. It adds to growing friction between Wikimedia and major AI developers over scraping and access.
A post on Schneier on Security examines a possible vulnerability in Apple’s automatic reboot feature. The issue may affect how Apple devices restart after inactivity, potentially undermining a security protection. It matters because automatic reboots can help keep encrypted data safer from certain attacks.
Security researchers scanned 15,465 publicly exposed Model Context Protocol (MCP) servers, the connectors that let AI assistants reach external tools and data, and found a range of misconfigurations and vulnerabilities. The findings highlight how quickly AI agent infrastructure has been deployed without proper hardening. As MCP adoption grows, exposed servers could give attackers access to credentials, internal systems, and the tools those agents control.