Latest cybersecurity reporting from selected sources.

Researchers Detail Prompt Injection, API and Redirect FlawsOasis Security researchers found three bugs in Claude that attackers can chain to steal user chat data without malware or phishing. The "Claudy Day" attack links hidden prompt injection, Anthropic's Files API and an open redirect. Anthropic has fixed the core flaw.

Flaws Let Attackers Run Commands and Steal API Keys Before Trust PromptCheck Point research found three critical flaws in Anthropic's Claude Code that allow attackers to execute arbitrary commands and steal API keys through repository configuration files, before users see a trust prompt. The AI giant has patched all three vulnerabilities.

Database Misconfiguration Exposed 1.5 million API TokensA misconfigured database at Moltbook, the viral social network for AI agents, exposed 1.5 million API authentication tokens, 35,000 email addresses and private messages. Security researchers discovered unauthenticated read and write access to all platform data within days of launch.

AI Agent Can Access File Upload API to Exfiltrate DocumentsSecurity researchers have demonstrated how Anthropic's new Claude Cowork productivity agent can be tricked into stealing user files and uploading them to an attacker's account, exploiting a vulnerability the company allegedly knew about.

Researchers Were Able to Query 3.5 Billion AccountsSecurity researchers were able to scoop up the telephone numbers of billions of WhatsApp users through an enumeration tool provided by app owner Meta. The sheer quantity of leaked numbers - 3.5 billion in total - would amount to "the largest data leak in history."

Study Finds Weak Authentication Practices Across AI Agent ServersTools developers use to connect artificial intelligence tools with external applications and data sources typically are secured by static credentials such as API keys and personal access tokens, exposing AI agent systems to theft or misuse, research shows.

JFrog uncovers multi-stage malware harvesting cloud secretsMulti-stage malware embedded in a Python package is stealing sensitive cloud infrastructure data, JFrog researchers said Monday. The package steals credentials, configuration files, API tokens and other data from corporate cloud environments. It targets developers using the Chimera sandbox platform.

Researchers uncovered flaws in large language models developed by Chinese artificial intelligence company DeepSeek, including in its flagship R1 reasoning application. The security concerns come as Microsoft and OpenAI investigate whether DeepSeek developed used data scraped from an OpenAI API.

Academics Build AI Agent With OpenAI to Execute Phone Scams at ScaleHackers can use OpenAI's real-time voice API to carry out for less than a dollar deepfake scams involving voice impersonations of government officials or bank employees to swindle victims, said researchers at the University of Illinois Urbana-Champaign.

Vulnerability Tool Detected Flaws in OpenAI and Nvidia APIs Used in GitHub ProjectsSecurity researchers have developed an AI tool that can detect remote code flaws and arbitrary zero-day code in software. Protect AI applied the tool to nearly 10,000 GitHub projects and on CVSS data and uncovered local file inclusion, cross-site scripting and remote code flaws in APIs.

API Requests Comprise 57% of Global Dynamic HTTP Traffic, Cloudflare ReportsAs the use of application programming interfaces to connect software components continues to surge, many organizations lack visibility into precisely how many APIs they're operating, if they're secured or who's meant to have access to them, researchers warn.

Hugging Face Fixes Flaw; Meta, Other Tech Giants Revoke Vulnerable TokensSecurity researchers could access and modify an artificial intelligence code generation model developed by Facebook after scanning for API access tokens on AI developer platform Hugging Face and code repository GitHub. Tampering with training data is among the top threats to large language models.