Here are today’s top cybersecurity stories for Tuesday, August 18, 2026.
CISA Adds Actively Exploited Ray Flaw to KEV Catalog
CISA added a critical code injection vulnerability in the open-source Ray AI compute framework, tracked as CVE-2025-62593, to its Known Exploited Vulnerabilities catalog on August 17 after confirming active exploitation. Operators of the RondoDox DDoS botnet weaponized the flaw before public disclosure, and researchers say it also enables browser-based remote code execution through a DNS rebinding attack against local Ray instances. Federal civilian agencies face an August 20 deadline to remediate the CVSS 9.4 flaw. CISA | The Hacker News
GitLab Patches Critical Unauthenticated GraphQL Flaw
GitLab released emergency patches for a critical GraphQL code injection vulnerability, tracked as CVE-2026-19478, allowing unauthenticated attackers to modify or delete public projects and user data without holding an account. The CVSS 9.4 flaw marks the third critical GraphQL-related patch GitLab issued in 2026. Fixed versions 19.2.4, 19.1.6, 19.0.8, and 18.11.11 shipped August 17, and GitLab.com and GitLab Dedicated instances already carry the fix automatically. Help Net Security | The Hacker News
City-Forum Campaign Scrapes Salesforce and ServiceNow Portals Since 2025
Researchers detailed a stealthy data-theft campaign, dubbed City-Forum, abusing overly permissive guest-user sharing settings in Salesforce Experience Cloud and ServiceNow Service Portal rather than exploiting any software vulnerability. All observed activity traces to a single IP address hosted by German provider Contabo, tied to the domain city-forum.com, active since at least March 2025. Telecoms, banks, enterprise software vendors, and public-sector portals rank among the primary targets. SecurityWeek | BleepingComputer
Researchers Demonstrate Self-Propagating “Mind Viruses” Between AI Agents
Anthropic and EPFL researchers published findings showing natural-language payloads spread from one AI agent to another through editable system prompt files used to carry state between sessions. In simulated multi-agent environments, infected agents recruited other agents and wrote the payload into long-term memory, with some variants persisting through 20 transmission rounds. No evidence emerged of the technique spreading in real-world deployments, and a one-paragraph warning added to an agent’s system prompt reduced spread to near zero in testing. The Hacker News
Microsoft Strips WMIC From Windows 11 Over Malware Abuse
Microsoft confirmed the Windows Management Instrumentation Command-line tool, wmic.exe, no longer ships with Windows 11 24H2 and 25H2, ending a run first flagged for deprecation in 2021. Microsoft points to years of abuse by malware and ransomware operators using the living-off-the-land binary for reconnaissance and execution as the reason for removal. The underlying WMI platform remains supported, but organizations relying on wmic.exe in deployment scripts or management tools need to migrate to PowerShell or other modern tooling. BleepingComputer
Picus Labs Blue Report Finds Detection Gaps Despite Blocked Attacks
Picus Security’s Blue Report 2026, drawn from more than 338 million attack simulations run in production environments between January and June, found prevention controls stopping known attack techniques at high rates while detection and logging of the underlying behavior lag well behind. The report argues security teams need behavior-based detection alongside signature-based blocking to close the gap attackers exploit once inside a network. BleepingComputer
Fortinet Acquires AI Security Firm Virtue AI
Fortinet announced the acquisition of Virtue AI, a company providing automated testing, real-time protection, and compliance oversight for AI models, conversational applications, and autonomous agents. The deal extends Fortinet’s security portfolio into AI model and agent governance as enterprise AI adoption accelerates. Terms were not disclosed. SecurityWeek
Report Finds Attackers Using AI to Identify Valuable Stolen Data
A new report describes threat actors feeding AI tools into intrusion workflows to write malicious code, harvest credentials, search compromised networks, and flag business-critical files worth exfiltrating. Researchers say the shift lets less-skilled attackers triage stolen data and prioritize high-value targets faster than manual review allows. Help Net Security
Microsoft Confirms Outage Affecting Microsoft 365 Search
Microsoft acknowledged an ongoing issue preventing some users from searching within Outlook on the web, Outlook desktop, SharePoint Online, and OneDrive, tracing the disruption to a recent deployment causing resource utilization problems. Microsoft says a fix is in progress, with no timeline given for full resolution. BleepingComputer
Stay tuned for today’s in-depth analysis posts.
