Meta workers challenge AI-scored layoffs

Former Meta workers allege that performance scoring penalized protected leave and fed into a layoff process. They’re asking a court to intervene before their terminations take effect on July 22. The complaint hasn’t been adjudicated; monitoring, scoring, and final layoff authority remain distinct.

Publishers allege Google repurposed book-service access

The Guardian

Publishers allege that books obtained under limited permissions for Google’s book services were later used to train Gemini. Their US case seeks remedies tied to that Google-specific chain of custody.

xAI’s generation permits face scrutiny

Techmeme

Reporting collected by Techmeme describes allegations of unpermitted power generation at a specific AI facility. The dispute ties local permit enforcement to the broader costs of data-center demand.

Regional customers face a reported $6.3B increase

Techmeme

A reported $6.3 billion multistate increase quantifies the power costs facing regional customers. The linked coverage connects AI data-center demand with customer bills.

Governors consider shifting more grid costs to developers

Techmeme

Republican governors are considering proposals that would make data-center developers absorb more of the power-system costs they create. The proposal follows reported impacts on permits, nearby communities, and regional customer bills.

Tailscale fixes an SSH argument bug that could open a root session

Tailscale

Tailscale says insecure argument handling in Tailscale SSH could permit a root session in affected configurations. The vendor bulletin documents the mechanism and upgrade target; it doesn’t describe unauthenticated internet-wide root access.

ASML raises guidance as chipmakers expand AI capacity

CNBC

ASML raised guidance as chipmakers expanded AI capacity. The reported sales range and planned EUV and DUV capacity increases offer a direct demand signal from the sole supplier of advanced EUV lithography equipment.

PalmClaw moves the agent loop onto the phone

arXiv

PalmClaw runs the agent loop on the phone, where it keeps session state and memory and executes skills and tools. Model inference uses a remote API. Its authors report an 11.5% relative improvement in task success and a 94.9% reduction in completion time over the strongest baseline.

RunAnywhere releases an on-device benchmark runner

RunAnywhere

RunAnywhere released a benchmark runner for developers measuring on-device AI performance. Together with PalmClaw, it gives builders a framework for local execution and a separate tool for measuring performance themselves.