New OpenAI research examines how ChatGPT changes the mix of tasks workers take on across different roles. The findings suggest AI is expanding job boundaries by helping people perform work that previously sat outside their usual responsibilities.
AMD has split its 2028 Zen 7 EPYC roadmap into the Florence, Ferrara, and Fidenza families. The plan extends an annual cadence across CPUs, GPUs, networking, and rack-scale systems through 2030, with products increasingly organized around AI agent workloads.
NVIDIA is working with Cadence and Synopsys to optimize key electronic design automation applications for its Vera CPU. It is also deploying Vera internally to speed the increasingly complex work of designing future CPUs, GPUs, and AI systems.
Cogent Security introduced VR-1, a reasoning model trained to discover and validate attack paths in live enterprise environments. The company says it proved twice as many paths as other frontier models on its new IntrusionBench benchmark at roughly one-quarter of the cost.
Vultr is positioning open, composable infrastructure and its AMD partnership as alternatives to tightly integrated AI cloud stacks. Its strategy emphasizes global performance per dollar while supporting the sovereignty requirements of different countries.
The MedusaHVNC malware-as-a-service operation launches legitimate browsers inside hidden Windows desktops to evade detection. This gives attackers persistent, covert remote access while keeping their activity out of the victim's visible session.
GitHub has added a three-day Dependabot cooldown before automatically proposing newly released package updates. PyPI is also rejecting file uploads to releases older than 14 days, narrowing opportunities for attackers to poison established packages.
NVIDIA, Microsoft, SpaceX, IBM, and other companies have formed the Open Secure AI Alliance to build and share open-source AI security tools. The group argues that defenders need openly available testing and protection technology as attacks involving frontier models become more capable.
Moonshot AI has published Kimi-K3 on Hugging Face, making the new model available to the developer community. The release quickly drew significant attention and discussion among practitioners evaluating the latest generation of open models.