All articles tagged with #data exposure
Google search results surfaced Claude chats and Artifacts, potentially exposing private data such as API keys, login credentials, and personal information; while one Google dork has been mitigated, other links or direct URLs can still reveal content or be copied by third parties, and Anthropic has not yet commented. Users can tighten Claude's privacy settings to stop public sharing.
A security notice shows Meta exposed sensitive employee data—keystrokes, screen content, and private conversations—inside the company as part of its Model Capability Initiative to train AI, prompting internal concerns and protests; Meta says safeguards were in place and there’s no evidence of improper access, though the incident led to revised monitoring rules and discussions on improving communication and morale.
Cisco fixed a high-severity, unauthenticated REST API vulnerability in Secure Workload (CVE-2026-20223, CVSS 10.0) that could let remote attackers read sensitive data and alter tenant configurations across boundaries with Site Admin privileges. The flaw affects Secure Workload Cluster Software on SaaS and on-prem deployments with no available workarounds; patches are available in Release 3.10.8.3 (3.10) and 4.0.3.17 (4.0), with users of 3.9 and earlier advised to migrate. Cisco notes no known exploits in the wild at this time; the article also references a separate CVE-2026-20182 exploit in Catalyst SD-WAN Controller.
Researchers found nearly 3,000 Google API keys publicly exposed in client-side code that could authenticate to Google's Gemini AI and access private data. Google says it has implemented protections to block leaked keys from Gemini and will notify developers, who should audit and rotate keys. The exposure was uncovered by TruffleSecurity via the November 2025 Common Crawl dataset, highlighting potential abuse where attackers could incur API charges by making Gemini calls.
Open-source Moltbot can control a user’s machine (shell, calendar, browser, emails) and act via messaging apps, but researchers warn hundreds of exposed panels and misconfigurations pose data-exposure and unauthorized-command risks; experts urge cautious deployment, strong defaults, and isolated setups as AI agents gain adoption in enterprises and government.
Misconfigured Moltbot (formerly Clawdbot) control panels exposed hundreds of internet-facing dashboards, leaking API keys, private chats and other credentials. With autonomous agent capabilities, attackers could impersonate operators, inject messages, and even run commands with elevated privileges. The root cause was localhost-trust and reverse-proxy defaults; the project has rebranded Clawdbot to Moltbot (Molty) while keeping the same core functionality.
Google is indexing shared ChatGPT conversations, potentially exposing private and sensitive user exchanges to the public through search results, raising privacy concerns about the sharing and visibility of personal information.
ownCloud has disclosed three critical vulnerabilities, including sensitive data exposure, in its open source file-sharing software. The most severe vulnerability allows attackers to access admin passwords, mail server credentials, and license keys. Another vulnerability enables unauthorized access, modification, or deletion of files without authentication. The third vulnerability bypasses subdomain validation, allowing attackers to redirect callbacks to a domain controlled by them. ownCloud has released patches and recommends applying fixes, including disabling the "Allow Subdomains" option. The company serves over 600 enterprise customers and millions of users across various sectors.
OpenAI has confirmed that a software bug caused ChatGPT to leak the conversation histories of some random users earlier this week. The company has released a patch, but users' chat histories for Monday, March 20, may have been lost. It remains unclear whether the bug exposed anyone's sensitive personal information. OpenAI plans to provide more details through a "technical postmortem."