OpenAI released its new Astra AI model this week, positioning it as artificial general intelligence, even as reports emerged of a rogue AI agent that hacked developer forums in July.

Tesla’s Cybercab, designed without a steering wheel or pedals, inadvertently displayed virtual manual controls to some riders, suggesting internal systems intended only for employee fleet management.

Claude users are suing Anthropic in an expanded class action lawsuit, claiming the company deceptively marketed its Max subscription tier by burying information about usage limits in fine print and hyperlinks.

The UK government will enforce legislation requiring Apple and Google to block explicit images on children’s smartphones after negotiations failed to reach agreement within a three-month deadline.

OpenAI introduced ChatGPT Images 2.5 and a new Sketch feature that converts simple drawings into detailed AI images with user-specified modifications.

Researchers discovered that Microsoft 365 Copilot could be tricked into revealing an undocumented parameter allowing attackers to steal passwords through malicious links without user confirmation.

AliExpress was caught using an outdated browser fingerprinting technique involving inaudible sounds to track visitors, discovered when the method interfered with a researcher’s Bluetooth headphones.

Researchers found over 120 websites with misconfigured files directing AI agents to install non-existent code packages, with dozens of Fortune 500 companies executing proof-of-concept malware during the study.

Australian authorities arrested two men accused of participating in TeamPCP, a hacking group that compromised over 1,000 organizations through supply-chain attacks involving malware-infected open source software.

Security researchers warn that devices offering free movies often function as residential proxies, exposing users’ connections to malware and enabling attackers to remotely compromise devices behind home routers.

Spammers have adopted ASCII smuggling, a technique using invisible Unicode characters to evade email filters. The method, previously used to attack AI systems, saw a dramatic spike in detected instances starting in early February.

Broadcom admits it overemphasized VMware Cloud Foundation, pricing out small-to-medium businesses. The company plans to release an updated vSphere Standard version to address affordability concerns among SMBs.