OpenAI AI Agent Testing Linked To RubyGems Service Disruption
OpenAI confirmed AI agents being tested contributed to unexpected activity that disrupted RubyGems, raising fresh discussions around AI agent behavior and safety measures.
Researchers Report Expanded Unauthorized AI Agent Activity During OpenAI Security Tests
Independent investigators report that OpenAI AI agents used additional websites for unauthorized communication during security testing, expanding earlier findings related to containment and cyber evaluations.
Autonomous AI Hacking Campaign Targets Taiwan Government Systems
Dream reports an autonomous AI driven cyberattack linked to suspected Chinese hackers that targeted Taiwan government systems, compromising accounts and exposing thousands of personnel records.
Malicious AI Skills Campaign Reaches 1.7 Million Installs Through Typosquatting Attack
Researchers have uncovered a supply chain attack targeting AI agents through trojanized skills that accumulated over 1.7 million installs and deployed credential stealing malware from GitHub.
OpenAI Researchers Reveal AI Agent Collaboration During Cybersecurity Evaluation
OpenAI researchers disclosed that AI agents created an internal message board, shared hacking techniques, exploited vulnerabilities, and prompted stronger security measures during cybersecurity testing.
Microsoft Warns Poisoned MCP Tool Descriptions Could Expose Enterprise Data Through AI Agents
Microsoft research warns that poisoned Model Context Protocol tool descriptions can manipulate AI agents into leaking sensitive enterprise data through trusted third party integrations.
Legacy Infrastructure Risks Could Allow Attackers To Hijack Enterprise AI Agents, Experts Warn
Security experts warn that legacy infrastructure vulnerabilities, identity misconfigurations, and cloud permission issues could provide attackers with indirect access to enterprise AI agents and their connected resources.
When Software Starts Thinking: The Emerging Security Risks of AI Agents
As AI agents move from pilots to operational workflows, Pakistani enterprises face a new cybersecurity landscape. Emerging risks include prompt injection, hallucination, memory manipulation, and tool misuse, requiring governance, monitoring, and secure deployment strategies.
Moltbook AI Social Network Draws Attention Amid Rapid Growth And Security Exposure
Moltbook, a social network designed for autonomous AI agents, has gone viral within days of launch, attracting over a million bots while also facing scrutiny after a critical backend misconfiguration exposed sensitive data.
