2026-08-05
AI Briefings Daily Report.

Anthropic Forms AI Chip Design Team, Develops Custom Chips In-House ↗

Anthropic is building a team for designing its own custom AI chips. The Claude maker said it would co-design hardware and models to help its technology run faster and more efficiently.

SpaceX's first earnings report beats expectations but AI capital spending scares off investors, stock falls 10% ↗

Privacy Center Currently, only residents from GDPR countries and certain US states can opt out of Tracking Technologies through our Consent Management Platform. Additional options regarding these technologies may be available on your device, browser, or through industry options like AdChoices. Please see our Privacy Policy for more information.

AI Agents Are Still Unable to Conduct Open-Ended AI Research ↗

Early evidence from two case studies

View: large language models cannot"Jump" ↗

China Unicom: intelligent computing scale reaches 45 EFLOPS, this year's computing-related investment will exceed 17.5 billion yuan ↗

China Unicom stated that it has clarified a full-year capital expenditure of 50 billion yuan for 2026, of which computing-related investment exceeds 17.5 billion yuan, accounting for 35% of total investment.

CrowdStrike Teams Up with AWS to Launch International AI Security Challenge with Total Prize of 100,000 US Dollars ↗

In this online game, players will use techniques such as prompt injection to try to control AI agents weaponized by fictional adversaries, and stop their actions before attacks occur.

Report: StepFun establishes two strategic lines of large models and agent terminals, phone business to operate independently and enter overseas market ↗

Inside StepFun, two strategic lines are being pursued simultaneously: one centered on large language models, and the other on intelligent agent terminals, with smartphones currently representing only preliminary exploration. (Huxiu)

Cloudflare proposes Agent Access Model ↗

The Agent Access Model proposes a new architecture to secure task-scoped agents using strict identity brokering, continuous mediation, and stateful trust.

Cloudflare uses identity-aware analytics to catch runaway AI behavior ↗

Identity-aware AI Gateway is now in open beta. User Insights turns that traffic into a behavioral baseline for every person and agent, and flags insider risk the moment it appears.

Insta360 GO Ultra AI voice assistant officially announced delayed launch; the team is doing final polishing and optimization ↗

Insta360 officially announced today that the GO Ultra AI voice assistant will be delayed. The official said that the team is currently making full effort on final polishing and optimization, and once the new launch time is confirmed, we will announce it through official channels as soon as possible.

Anthropic Confirms for the First Time Its Self-Developed AI Chip: Custom-Built for Claude, Company Is Forming a Team ↗

Will adopt hardware-model co-design to improve Claude operation efficiency, and will still retain multi-supplier strategy in the future; OpenAI has previously started a self-developed AI chip project. #AI chip# #Anthropic Claude#

MacPaw Partners with Liquid AI to Provide On-Device Inference Capability for SetApp Developers ↗

MacPaw is building a local version of its AI assistant Eney using Liquid AI's models.

Warp Agent Command-Line Interface Released: Multi-Model Agent Usable in Any Terminal ↗

The Warp Agent is now available as a standalone CLI that you can use anywhere – Ghostty, iTerm 2, VS Code, built-in Windows and Mac Terminals – and provides a superior experience for developers who do heavy terminal workflows.

Microsoft for internal"AI Fever"Cooling down: Limiting token usage, GPT-5.6 set as default model ↗

Microsoft Executive Vice President Parikh emphasized that the company has no intention of slowing down its transformation into an 'AI-first' enterprise. As models and products continue to change, internal AI policies will also continue to adjust. 'Our goal is not to use fewer Tokens, but to make each Token work harder.'

Report: US frontier AI model cybersecurity review mechanism temporarily does not cover open-weight models ↗

The US government is not expected to publicly disclose the complete model review mechanism, but these contents are also not confidential, and enterprises that choose to participate can obtain relevant detailed information.

Google Assistant will gradually be retired starting next month,"Hey Google"Will be taken over by Google Gemini ↗

According to foreign media Android Authority reported today (5th), the migration work will start from September 4 and be completed in batches over the following weeks. After the device switch is completed, users will no longer be able to use Google Assistant through phones, tablets or paired devices, nor can they switch back to the original service.

When AI benchmarks stall: a systematic study on benchmark saturation ↗

Abstract page for arXiv paper 2602.16763: When AI Benchmarks Plateau: A Systematic Study of Benchmark Saturation

WindBorne Systems Completes $37 Million Series B Financing, Uses AI Weather Model to Make Weather Forecasting More Profitable ↗

WindBorne Systems has raised a $37 million Series B round to scale its weather balloons and AI forecasts.

U.S. appeals court overturns ban, Perplexity AI shopping agent returns to Amazon ↗

A US appeals court has overturned Amazon's injunction against Perplexity's AI shopping agents, ruling that it's the users who access Amazon, not the startup. It's the first federal appeals court decision on whether AI agents can lawfully act on online platforms on behalf of users, and it could reshape the entire AI agent industry.

Trump administration AI testing framework has limited scope and vague definition, completely excludes open-source models ↗

The Trump administration’s framework for assessing potential cybersecurity risks posed by advanced AI reportedly has no interest in testing open models.