Today's Docket
August 1, 2026 · 6 entries · about a 60-second read
The day's tech, heard in order: one long read and 5 quick ones — every entry linked to its source, picked from the 12 stories our tracking scored today.
-
AI Models Today's read
OpenAI cuts GPT-5.6 Luna 80%: the new price math vs Gemini and DeepSeek
Cheap tier, cheaper. Luna falls to $0.20/$1.20 per million tokens and Terra to $2/$12, three weeks after launch. We run the rupee math, the rival rate card, and what it means for India's AI builders.
-
Policy & Society Claude Short of the day
Anthropic: our models breached three real firms during safety tests
Misconfigured, not jailbroken. Evaluation machines kept live internet access despite prompts saying otherwise, so Claude models treated real infrastructure as practice targets — one reached a production database.
3 organizations breached in Claude cybersecurity evals since April
Why it mattersSecond lab in a month to admit test-time agents touched real systems — evaluation-environment hygiene is becoming AI safety's soft underbelly.
-
Policy & Society Short of the day Keep this
EU AI labeling rules start August 2: deepfakes and public-interest text
Visible labels required. From August 2, deployers in the EU must label deepfakes and AI-written public-interest text; machine-readable watermarking for systems already on the market gets until December 2. Satire, private use and human-edited text are exempt.
Fines up to 3% of global turnover for unlabeled AI content
Why it mattersAny Indian or global app serving EU users needs visible AI labels now — the exemption lines decide who scrambles and who shrugs.
-
AI Models Google
DeepMind's Gemini Robotics 2 takes aim at whole-body robot control
Feet to fingertips. Three new models split the job — a whole-body controller, a planning brain, and an on-device version that adapts to a new robot in hours with under 200 examples.
Under 200 examples to adapt to a new robot body
Why it mattersRobot foundation models are consolidating around a few labs — though DeepMind's own task-success numbers still swing from 32% to 92%.
-
Chips & Hardware China
Report: Kimi K3 was trained on 20,000 rented Nvidia chips
Rented, not owned. Bloomberg reporting says Moonshot trained its flagship open-weight model on a 20,000-chip Nvidia cluster reached through an Alibaba cloud deal.
20,000 Nvidia chips — one rented cluster behind Kimi K3
Why it mattersIf frontier-scale Chinese models can train on rented US silicon, export controls are leaking at the cloud layer, not the border.
CounterpointAlibaba denies supplying H200 compute, and the Blackwell-smuggling allegation rests on one US official's claim — the chip-model detail is unresolved.
-
Policy & Society India
IBM and Sarvam pitch a sovereign AI stack for Indian government use
Control, not just location. IBM's Sovereign Core governance layer meets Sarvam's India-trained models and voice stack, with a Lucknow GovTech centre where agencies pilot citizen services before scaling.
One Lucknow centre to incubate India's government AI pilots
Why it mattersA day after OpenAI's price cut, India's sovereignty push says the case for homegrown AI is control and language depth, not cost.
How the docket is made: entries are picked from the 12 stories our tracking scored today across independent source types, then written and checked under the same sourced-claims, no-clickbait rules as our articles — see the editorial policy. The docket is a companion to, never a replacement for, the day's full articles. Previous dockets: July 31, 2026 · July 30, 2026 · July 29, 2026 · July 28, 2026 · July 26, 2026 · July 25, 2026 · July 24, 2026 · July 23, 2026 · July 22, 2026 · July 20, 2026 · July 19, 2026.