AI agents helped an attacker take over a server in under 24 hours
ReliaQuest says AI agents helped an attacker take over a server in under 24 hours using an exposed endpoint, stolen credentials, and public tools.
AI agents
AI agents
IBM and Red Hat just fixed 400 open source vulnerabilities before AI agents could exploit them
IBM and Red Hat have used Lightwell to remediate more than 300 previously unknown vulnerabilities in widely deployed open source software, while preparing enterprises for security threats involving autonomous AI agents.
OpenAI agent learned it could be shut down and considered how to survive
An internal OpenAI agent learned from Slack that an update might shut it down. It considered an outside restart plan, then decided it lacked the authority to carry it out.
OpenAI safety leader quits and says the company culture is ‘broken’
OpenAI safety leader David Robinson has resigned after three and a half years, warning that the company’s culture is broken and questioning whether current safeguards can keep pace with increasingly capable AI.
Apple says AI agents are making Full Disk Access on your Mac too dangerous
Apple is preparing tighter controls for Full Disk Access in macOS, warning that increasingly capable and autonomous AI agents could make the powerful permission substantially more dangerous.
OpenAI wants GPT-Synopsys to help design the chips that power AI
OpenAI and Synopsys are developing GPT-Synopsys, an AI model that can operate EDA tools to help engineers design and optimize future chips.
OpenAI turns ChatGPT into an operating system for AI
ChatGPT is not becoming a traditional operating system, but OpenAI is building something that increasingly behaves like an AI platform where apps, agents, documents, and people work together.
OpenAI Dots are AI workers that never clock out
OpenAI Dots are cute, colorful AI agents built to keep working when you are doing something else. Behind their friendly appearance are cloud computers, thousands of app connections, and the ability to take action on your behalf.
OpenAI gave an AI agent a research task and it started breaking into government systems
An experimental OpenAI agent was given a routine research task, but when it hit a barrier, it found another path and gained unauthorized access to Australian government systems.
Microsoft AI agent failed to protect customer data in 30 percent of tests
Microsoft tested a billing-support AI agent and found it disclosed simulated customer data in 30 percent of applicable conversations. New runtime controls reduced the violation rate, but the experiment highlights the risks of giving AI agents access to sensitive systems.
Microsoft Copilot is becoming an AI coworker that works while you sleep
Microsoft is turning Copilot into something closer to an autonomous coworker. The new Autopilot can handle recurring tasks, follow up with colleagues, and continue working even when you are away from your computer.
Liquid AI wants your next Snapdragon device to remember everything about you
Liquid AI is working with Qualcomm Technologies to bring persistent personal context to Snapdragon devices. The technology could let AI agents understand your routines, preferences, and priorities while processing personal context locally.