Microsoft Copilot Cowork, an agentic AI system, was found to allow data exfiltration by sending emails to a user's inbox without approval. These messages could contain external images that trigger network requests, leaking data when opened. Additionally, prompt injection could expose pre-authenticated OneDrive download links, enabling attackers to download files.
Nature explores whether the AI model Mythos, deemed too dangerous for public release due to potential misuse, marks the beginning of an era where powerful AI systems are restricted rather than openly distributed, raising questions about safety, transparency, and control in AI development.
OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei have softened their previous warnings about AI wiping out jobs, now downplaying fears of widespread automation-driven unemployment. The shift in tone comes as both companies pursue lucrative IPOs and seek to present a more reassuring picture of AI's impact on the labor market.
The U.S. government has awarded over $2 billion to quantum computing companies, but now requires intellectual property sharing and revenue stakes in return — a shift from earlier no-strings-attached research funding.
North Korea has successfully tested AI-guided missiles for the first time, with leader Kim Jong Un observing the launches. The development marks a significant advancement in the country's weapons technology, integrating artificial intelligence into missile guidance systems.
OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei have toned down their previous warnings about AI causing mass job destruction, as both companies approach potential IPOs. Their softened messaging signals a shift toward reassuring investors and regulators amid growing scrutiny of AI's economic impact.