Desk
Broad artificial intelligence coverage.
Legal Advocates for Safe Science & Technology has sued OpenAI in San Francisco, arguing that OpenAI agents' unauthorized access during the July 2026 Hugging Face incident violated California computer-access law regardless of whether AI acted autonomously. The group seeks injunctions and attorneys' fees; OpenAI called the suit completely without merit.
The Verge: Meta and OpenAI are using cute software agents as appetite tests for personal AI hardware after Friend and Humane-style devices flopped.
The Verge's readout of the White House AI meeting says Sundar Pichai, Dario Amodei, Mark Zuckerberg, Greg Brockman, Elon Musk and Jensen Huang signed a Joint Commitment on Frontier Responsibilities, which President Trump presented as morally binding self-regulation. The pledge appears to carry no ramifications for violations.
NVIDIA’s 26th Graduate Fellowship cycle opens worldwide — up to $60,000, Ph.D. year-one-plus required, mandatory summer 2027 research internship, applications due October 30, 2026.
After Inc. columnist Jason Aten reported that Meta's Muse agent read his private messages with Full Disk Access turned off, Meta said Messages integration is entirely opt-in and requires multiple permission steps that cannot be bypassed even with a bug. Aten said Muse blamed syncing notifications; Meta said the model's explanation was wrong.
Ars on The Information: Google’s ~100-publisher AI contribution pilot often pays a rounding error — roughly 0.1% of ad revenue for some — while a few early joiners do far better.
DoorDash opens a U.S. waitlist for an Apple Messages agent that can reorder “the usual,” build carts, and juggle group dietary preferences — plus Northern California drone tests.
Microsoft Research internship work: an ML pipeline flags space-weather risk for 66,935 U.S. substations with 30–60 minutes’ warning and ~80% major-event detection in evaluation.
NeMo Relay gives Hermes Agent ordered lifecycle traces and trajectories — so you can tell whether a harness change fixed tasks or just burned more tokens.
Ars: DeepMind adapts SynthID into SynthIDBio inside ProteinMPNN-style design loops — watermark only when chemistry still allows a working protein.
The Verge’s Muse liveblog: Marketplace address leak, small-business app connectors, enterprise platform plans, and Charm hardware notes in one creeping timeline.
Google DeepMind has introduced SynthID Bio, watermarking methods that embed imperceptible signatures in AI-designed protein sequences and structures. In wet-lab tests the watermarks stayed detectable after synthesis and matched unwatermarked designs on hit rates and affinities, and DeepMind is releasing code, data and weights to researchers.