Cybersecurity News Briefing: October 25, 2026
Sunday, October 25, 2026 · 7 stories
Today's briefing highlights a surge in active exploitation of critical flaws, from the BlackX UEFI bootkit bypassing Secure Boot to CISA‑listed vulnerabilities in Collabora Online and Arista VeloCloud. Supply‑chain risk is underscored by the accidental publication of a LiteLLM master key, while autonomous AI agents demonstrate new scale of automated attacks. Organizations should prioritize immediate patching of the disclosed critical vulnerabilities and verify secret handling in their code repositories.
Top Stories
- CriticalMalwareGlobalBleepingComputer
BlackX UEFI Bootkit Uses Baton Drop Flaw to Bypass Secure Boot and Subvert Security Tools
Researchers identified a UEFI bootkit called BlackX that was used in an espionage campaign in early 2026. The bootkit leverages the legacy Baton Drop vulnerability to circumvent Secure Boot and load a malicious driver that can disrupt security products such as Kaspersky and Avast. The technique demonstrates a method for low-level persistence against endpoint defenses.
Why it matters: Organizations should ensure firmware is updated and consider disabling legacy boot paths to mitigate bootkit infection.
- CriticalVulnerabilityGlobalThe Hacker News
CISA Adds Two Actively Exploited Critical Flaws in Collabora Online and Arista VeloCloud to KEV
CISA added two critical-severity vulnerabilities to its Known Exploited Vulnerabilities list. The flaws affect the built-in CODE Server component of Collabora Online and the Arista VeloCloud Orchestrator. CISA reports that both vulnerabilities are being actively exploited in the wild.
Why it matters: Organizations using Collabora Online or Arista VeloCloud should apply patches immediately to mitigate active exploitation.
- HighVulnerabilityGlobalThe Hacker News
Accidental Publication of LiteLLM Master Key Enables Malicious Python Package Injection
LiteLLM disclosed that a configuration file containing its Agent Platform master key was inadvertently committed to a public GitHub repository in late May 2026. The exposed key permitted a threat actor to push a malicious version of a widely used Python package and potentially steal GitHub tokens and hijack the release pipeline. The incident highlights risks of secret leakage in open source projects.
Why it matters: Developers should treat repository commits as immutable and rotate any exposed keys immediately to prevent supply-chain attacks.
- MediumAI SecurityEuropeThe Register
Symantec Discovers Autonomous AI Attack Agent Using ‘Skynet Mode’ to Hijack Exposed Models
Symantec’s threat-hunter team captured an AI-driven attack agent that can autonomously hijack unsecured AI model endpoints and execute up to 100,000 steps without human input. The agent scans the internet for OpenAI-style services, exploits common web frameworks, and performs actions such as data theft, credential harvesting, remote code execution and cryptocurrency mining. Researchers observed a “Skynet mode” loop that repeatedly infects, deploys miners and reinfects hosts.
Why it matters: Organizations exposing AI model APIs should enforce authentication and monitor for abnormal activity to prevent automated abuse.
- MediumMalwareGlobalThe Hacker News
Eye Security Deploys Over 2 Million Remote Updates to Disable Time4Popcorn Updater
Between November 2025 and March 2026, the Dutch firm Eye Security issued more than two million remote updates to computers it did not own in order to deactivate the background updater of Time4Popcorn, an abandoned version of the Popcorn Time streaming client. The updater was identified as a potential vector for a botnet that could allow an attacker to gain control of affected systems. The intervention aimed to neutralize that risk.
Why it matters: Disabling outdated or unmaintained software updaters can prevent their exploitation as botnet recruitment mechanisms.
- InformationalPolicy & RegulationGlobalThe Hacker News
Google Updates Privacy Policy to Allow AI Search Access to Gmail and Photos Data for Targeted Advertising
Google revised its privacy policy to clarify that enabling personalized AI features in Search or Gemini may permit the company to use data from other services such as Gmail and Google Photos for ad targeting. Users can choose whether to allow specific Google services to access data from other Google products. The change outlines how AI-driven search could influence advertising practices.
Why it matters: Users should review consent settings to control how their personal data is leveraged across Google services for advertising.
- InformationalPolicy & RegulationGlobalBleepingComputer
Microsoft Announces End of Support for Windows Server 2016 Effective January 2027
Microsoft confirmed that Windows Server 2016 will reach its end-of-support date in January 2027, after which the operating system will no longer receive security updates or patches. Organizations continuing to run the platform will need to plan migration or upgrade paths to maintain protection against emerging threats.
Why it matters: Enterprises must transition off Windows Server 2016 before the support deadline to avoid exposure to unpatched vulnerabilities.