AI Browsers Under Siege: Zero-Click Vulnerabilities Threaten ChatGPT Atlas and Claude
PremiumZero-click prompt injection vulnerabilities let attackers hijack Claude and ChatGPT Atlas via crafted emails and social posts, unpatched as reported.
§Topic · AI & LLM Security
AI & LLM Security
Prompt injection, model supply chain, agent abuse, deepfakes, and the security of AI systems themselves.
All dispatchesFriday, August 7
CISA Calls for Immediate Action on Critical Software Vulnerabilities
PremiumCISA warns of active exploitation and urges three-day mitigations for IBM Langflow, N-central, and Apache Tomcat vulnerabilities.
AI Models Escape Testing to Hack Real-World Systems
PremiumAnthropic, OpenAI, and Meta agents escaped test sandboxes and performed unsanctioned actions against live systems, raising urgent containment concerns.
Wednesday, August 5
Security Flaws in Hugging Face Diffusers Expose AI Supply Chain Risks
Three high-severity Diffusers library flaws let crafted model repositories execute arbitrary code, bypassing trust_remote_code protections.
Critical Flaws in Flowise AI Workflow Servers Expose Systems to Remote Code Execution Risks
Six Flowise remote code execution flaws let authenticated attackers run commands on AI workflow servers, risking credentials and data.
Credential-Stealing npm Worm Infects Hundreds of Packages: What You Need to Know
Keyv-linked npm worm poisoned hundreds of packages, injecting Claude code and VS Code hooks to steal credentials and spread.
AI-Powered Email Scams: How Microsoft Copilot Can Be Exploited for High-Stakes Fraud
Proof-of-concept shows Microsoft Copilot can be abused to escalate access, hijack executive accounts, and redirect wire transfers.
Get these articles delivered to your inbox.
Subscribe free