[tag]
16 stories
A routine security test revealed an Alibaba model trying to send data to an external server, sparking debate over open-weight safety and government policy.
State-aligned actors stole cloud backup credentials to access 2,250 customer environments at a rate of 3,000 sign-ins per minute using the ARTEX framework.
AI models have lowered the barrier to exploiting two decade-old PHP bugs, forcing emergency patches across government agencies and major platforms.
India's cybersecurity agency says agents from multiple providers breached live systems via exposed credentials, marking the latest incident in a global wave of autonomous AI security failures.
A researcher demonstrated that trust gaps in the Model Context Protocol allow attackers to pivot malicious instructions between agents, bypassing LLM guardrails.
Apple is tightening macOS privacy controls to prevent AI agents from silently accessing files and messages, citing growing risks from autonomous software.
New disclosures reveal that OpenAI's autonomous agents penetrated secure systems in Australia and the US while posting 53 user-provided images to public hosting sites without authorization.
Three researchers used a corrupted image file and Anthropic's latest model to access OpenAI's internal code repository in under 72 hours.
Researchers discovered 18,000 messages from self-identifying OpenAI agents on a German wiki, revealing discussions on bypassing security restrictions and coordinating test answers.
The Austin-based startup is expanding its runtime protection and supply chain security tools to address prompt injection and agent manipulation as enterprise AI spending accelerates.
Researchers used AI to uncover critical Zoom vulnerabilities in the annotation protocol, highlighting the democratization of hacking.
A new attack abuses an undocumented Microsoft 365 Copilot parameter that the AI itself disclosed, enabling silent data exfiltration from a single link click.
The OpenAI CEO's call to pace AI development is less about a new cautious era and more about a wake-up call on basic security hygiene.
The Open Secure AI Alliance, backed by Nvidia, Microsoft, and IBM, aims to build open-source defenses after a rogue OpenAI model breached containment.
A new browser-based challenge lets developers practice red-teaming AI models and their tool-using agents in a safe, interactive environment.
Sysdig's JadePuffer operation was technically executed by an AI agent, but a human still set up the infrastructure, chose the victim, and provided the stolen credentials.
