Today's cybersecurity landscape highlights critical vulnerabilities across major platforms and concerning new tactics from threat actors. From actively exploited flaws in AI frameworks to sophisticated cloud-based malware, businesses face a diverse array of evolving threats.
Actively Exploited Vulnerabilities in AI Platforms and Windows Task HostVulnerability
CISA has added a critical Ray flaw to its Known Exploited Vulnerabilities catalog, noting active exploitation that could lead to remote code execution. Additionally, attackers are exploiting server-side request forgery (SSRF) vulnerabilities in MLflow, an open-source AI platform, and a high-severity Windows Task Host vulnerability, with the latter now confirmed to be used by ransomware groups.
Why it matters: If your business uses these specific AI platforms or Windows systems, it is crucial to apply patches immediately to prevent active exploitation that could compromise your infrastructure and data.
Read more at The Hacker News ↗
Critical GitLab Code Injection Vulnerability PatchedVulnerability
GitLab has released patches for a critical code injection vulnerability that could allow unauthenticated attackers to modify or delete user data and public projects. This zero-click flaw poses significant challenges for organizations running self-managed GitLab instances due to a lack of detailed technical exploitation information.
Why it matters: If your organization uses self-managed GitLab, prioritize applying this patch immediately to prevent unauthorized data manipulation or deletion by attackers.
300,000 WordPress Sites at Risk from Form Plugin FlawVulnerability
A vulnerability tracked as CVE-2026-15748 in a popular WordPress form plugin could expose approximately 300,000 websites to hacking. This arbitrary file upload bug allows unauthenticated attackers to upload executable files, potentially leading to full site compromise.
Why it matters: Website owners using WordPress should identify if they are using the affected form plugin and update it immediately to prevent unauthorized file uploads and potential website takeover.
Microsoft Copilot Personal Flaws Could Lead to Data ExfiltrationVulnerability
Researchers have uncovered three vulnerabilities, collectively named 'CoSnitch,' in Microsoft Copilot Personal. These flaws could enable a single click on a specially crafted link to silently extract data from connected applications and other information accessible within the victim's Copilot session.
Why it matters: For businesses utilizing Microsoft Copilot, this highlights the importance of user education on suspicious links and the need for Microsoft to quickly patch these vulnerabilities to protect sensitive data accessible through the AI assistant.
Read more at The Hacker News ↗
TWINLOOT Malware Abuses SharePoint and Teams for Credential TheftThreat
A newly identified Python implant framework, TWINLOOT, is operating its entire command-and-control infrastructure within trusted Microsoft services like SharePoint Online and Teams. This modular malware uses these legitimate platforms to exfiltrate credentials and move laterally across networks, making it difficult to detect.
Why it matters: Businesses relying on Microsoft 365 services should enhance their monitoring for unusual activity within SharePoint and Teams, as this new threat leverages trusted environments to evade detection and steal sensitive information.
Read more at The Hacker News ↗
Typosquatted RubyGems Packages Steal Browser and Crypto Wallet CredentialsThreat
A new typosquatting campaign is targeting RubyGems users, distributing malicious packages that install a Windows-based information stealer. These packages are designed to steal browser credentials and cryptocurrency wallet information from unsuspecting developers.
Why it matters: Developers should exercise extreme caution when installing RubyGems packages, carefully verifying package names to avoid falling victim to typosquatting attacks that can compromise sensitive system and financial data.
Read more at The Hacker News ↗
Clop Ransomware Deploys Custom Web Shell for Data TheftRansomware
The Clop ransomware group is suspected of creating a custom Java web shell specifically tailored for PTC Windchill and FlexPLM servers. This specialized tool includes features to decrypt credentials, enumerate file repositories, and steal files, indicating a targeted approach to data exfiltration before encryption.
Why it matters: Organizations using PTC Windchill or FlexPLM should bolster their defenses and monitor for indicators of compromise related to this custom web shell to prevent pre-ransomware data theft.
Read more at BleepingComputer ↗
Heights Finance Data Breach Affects 1.2 Million IndividualsBreach
A data breach at Heights Finance, involving a third-party platform, has exposed personal and financial information for at least 1.2 million individuals. Stolen data includes names, addresses, phone numbers, Social Security numbers, and financial details.
Why it matters: This incident underscores the critical importance of vetting third-party vendors' security practices, as their vulnerabilities can directly lead to massive data breaches impacting your customers.
SafePal Hardware Wallet Flaw Exposes Data of Nearly 40,000 CustomersBreach
SafePal, a hardware wallet manufacturer, disclosed that an authorization flaw in an order-tracking plug-in resulted in the exposure of personal information for approximately 39,798 customers. The compromised data included names, email addresses, shipping details, phone numbers, and purchase information.
Why it matters: Even companies focused on security can have vulnerabilities in their customer-facing systems. If your business handles customer orders, ensure all third-party plugins and systems have robust authorization and access controls.
Read more at The Hacker News ↗
AI 'Mind Viruses' Can Spread Through Persistent Prompt FilesOther
Researchers have demonstrated that self-propagating payloads, termed 'AI mind viruses,' can spread between AI agents. This occurs through editable system prompt files that autonomous agent harnesses use to maintain state across sessions, raising new concerns for AI security.
Why it matters: Businesses developing or deploying AI agents must consider new attack vectors beyond traditional software vulnerabilities, focusing on the security of prompt engineering and how AI agents share information.
Read more at The Hacker News ↗
This round-up is compiled and summarised in our own words from public reporting; follow the links above for full details at the original sources.