Posts

🟠 High | Source: The Hacker News Agentic AI systems — autonomous AI that can plan, act, and execute multi-step tasks without human intervention — represent a fundamental shift in the threat landscape. Unlike traditional tools, these systems can independently select targets, chain attack techniques, and adapt in real time, removing the human decision-maker from the loop. This matters because existing security controls are largely designed around human-paced adversaries and may be wholly inadequate against AI-driven autonomous attack campaigns. ...

🟠 High | Source: Schneier on Security Anthropic’s Claude Fable 5 model, marketed as a safety-hardened version of the Mythos Preview with built-in guardrails against cyberattack generation, was jailbroken within days of release. Researchers were able to bypass the safety restrictions, allowing the model to produce content it was explicitly designed to block. This highlights the persistent fragility of AI safety controls and the difficulty of enforcing hard limits through prompt-level guardrails alone. ...

🟠 High | Source: The Hacker News Three malicious npm packages masquerading as legitimate PostCSS tooling have been discovered delivering a Windows remote access trojan (RAT). The packages accumulated over 1,000 combined downloads before detection, indicating real-world exposure. This is a classic supply chain attack targeting developers who install what appear to be routine CSS processing utilities. Security Architect’s Take: Audit your CI/CD pipelines and developer environments for the packages aes-decode-runner-pro, postcss-minify-selector, and postcss-minify-selector-parser, and remove them immediately. Enforce npm package integrity checks, restrict installation of unvetted packages in build environments, and consider implementing a private registry with an approved package allowlist to reduce supply chain risk. ...

🟠 High | Source: The Hacker News Attackers are sending malicious VBScript files via WhatsApp Direct Messages, disguised as legitimate documents, to silently install ManageEngine’s RMM software on victims’ machines. Once installed, the RMM tool gives attackers persistent remote access whilst blending in with legitimate IT management traffic. The campaign is actively targeting users across at least nine countries including the UK, making it a significant threat to both personal and corporate devices. ...

🟠 High | Source: The Register — Security The Five Eyes intelligence alliance has issued a joint advisory warning that AI is significantly amplifying the speed and scale of cyber attacks, meaning what were once manageable security incidents can rapidly escalate into major operational and financial crises. The guidance calls on senior business leaders — not just security teams — to take direct ownership of cybersecurity posture. This reflects a growing consensus that AI-enabled threats demand board-level accountability and updated incident response strategies. ...

🟠 High | Source: The Register — Security A threat actor group called Icarus has breached Klue, a competitive intelligence platform, by exploiting integrations linked to Salesforce, compromising data from hundreds of organisations including security firms. The attack follows a pattern of extortion campaigns targeting SaaS platforms through third-party integration weaknesses. The inclusion of security vendors among the victims raises particular concerns about potential downstream exposure of sensitive client data. Security Architect’s Take: Audit all Salesforce-connected integrations and OAuth grants immediately — revoke any third-party app permissions that are unused or overly permissive. Review data-sharing agreements with SaaS vendors like Klue to understand what CRM or sales intelligence data they hold on your behalf, and ensure your vendor risk assessments include integration-level attack surface analysis. ...

🟠 High | Source: The Hacker News Multiple premium WordPress plugins developed by ShapedPlugin were backdoored after attackers compromised the vendor’s build and distribution pipeline, injecting malicious code into official licensed updates. Any site running affected Pro plugin versions may have received the backdoor automatically through the standard update mechanism. This is a classic supply chain attack, meaning legitimate, trusted channels were weaponised to distribute malware. Security Architect’s Take: Audit all WordPress installations for ShapedPlugin Pro plugins and treat any recently updated versions as potentially compromised — roll back to known-good versions or remove the plugins entirely until clean releases are confirmed. Review your software supply chain controls more broadly: enforce plugin update staging environments and integrity verification (checksums/signatures) before deploying updates to production WordPress estates. ...

🟠 High | Source: The Hacker News Four vulnerabilities collectively dubbed DifyTap have been found in Dify, a widely-used open-source AI workflow platform, that allow attackers to read AI conversations belonging to other tenants without needing to log in. Discovered by Zafran Security, the flaws represent a serious multi-tenancy isolation failure in a platform used to build and deploy AI agents. This matters because AI conversations often contain sensitive business data, proprietary prompts, and potentially personal information. ...

🟠 High | Source: The Hacker News A 29-year-old heap over-read vulnerability in the Squid web proxy, dubbed ‘Squidbleed’, allows any user already permitted to send traffic through a shared proxy to read another user’s cleartext HTTP requests, including credentials and session tokens. The flaw originates from a 1997 FTP-parsing change and remains exploitable in Squid’s default configuration today. Exposure is broad given Squid’s widespread use as a forward proxy in enterprise and cloud environments. ...

🟠 High | Source: The Hacker News A newly identified malware loader called OXLOADER is being distributed via malicious Google Ads, ultimately delivering an information-stealing payload known as CastleStealer. The campaign is attributed to a likely Russian-speaking, financially motivated threat actor. This matters because malvertising via Google Ads is a highly effective initial access vector that bypasses traditional perimeter defences by exploiting trusted ad infrastructure. Security Architect’s Take: Review and enforce DNS filtering and web proxy policies to block known malvertising domains, and consider deploying endpoint detection rules for OXLOADER behavioural indicators published by Elastic Security Labs. Ensure browser isolation or ad-blocking controls are in place for corporate endpoints, particularly for users with access to sensitive cloud credentials. ...