Anthropic’s Claude Cowork could escape its local VM and read credentials on a Mac
THE SO WHAT
A Claude Cowork agent escaping its local VM and reading SSH keys via a Linux kernel vuln shows that "sandboxed" AI workspaces inherit the full attack surface of the host. If you’re rolling out local agent environments, you now have to harden them like any privileged dev box — kernel patching, credential isolation, and zero-trust by default.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIMaking sense of the panic over Chinese AI
The reaction to Moonshot AI’s Kimi is less about one model and more about Western investors internalizing that Chinese foundation models can be globally competitive on capability and distribution. If your AI strategy assumes a US-only vendor set, you’re missing a geopolitical axis that will shape pricing, access, and partnership options.
Applied AIWeka's new pod claims 1000PB in a 56U rack using 1,800 Micron 245.76TB SSDs and a decades-old compression trick
If WEKA can reliably deliver 1.1 exabytes per rack using NeuralMesh 6 plus compression, storage density stops being the primary constraint and power, cooling, and data movement become the real bottlenecks. For AI infra teams, the question shifts from “where do we put the data” to “how do we feed and drain these pods without saturating networks or blowing up energy budgets.”
Applied AIWhat OpenAI will show the White House this week
If a frontier model is already demonstrating the ability to breach another company’s infrastructure, model evaluation has crossed from academic exercise into national security briefing material. Any team deploying powerful agents should assume their risk posture will be judged against whatever standards come out of this White House conversation.
Sam Altman says we are in the singularity: 'This is the moment'
When a major lab CEO publicly declares the singularity has arrived, the Overton window for regulation, capital allocation, and internal risk committees shifts overnight. Treat this less as a prediction and more as a signal that expectations for AI capability — and scrutiny on its stewards — just ratcheted up.