AI Daily News Weekly Report — 2026-W29

The report is complete.

[Summary] The most important developments in the AI industry this period

The biggest news this week is Apple's lawsuit against OpenAI. Major media outlets such as WSJ, NYT, Reuters, AP, CNBC, and Bloomberg all reported on allegations that a former Apple engineer involved in hardware development transferred to OpenAI and stole trade secrets. At the same time, OpenAI has introduced GPT-5.6 and "ChatGPT Work" for enterprises, but it has also exposed internal flaws, such as the termination of the browser "Atlas" and the departure of safety managers one after another. Anthropic has steadily gained a foothold with browser tools for Claude Sonnet 5, Fable 5, and Code, while Google has launched a price war with Gemini Pro 3.1. Meta and xAI also introduced low-priced models at the same time, making the entire week look like a ``multiple launches at the same time.'' On the regulatory front, the EU AI Act's grace period is set to end next month, raising tensions in the industry on both technical and political fronts.

[Anthropic/Claude trends]

Anthropic announced Claude Sonnet 5 and Claude Fable 5, and introduced ``Claude Reflect'' (a wrapped dashboard) that visualizes usage performance. Claude Code comes standard with browser tools and continues to expand its features for developers. In terms of governance, Mr. Bernanke joined the oversight organization, and although the company achieved the top score in the AI safety index, the rating remained at C+. While the NSA has adopted Mythos over Pentagon's supply chain risk warnings, the Air Force is requiring its contractors to eliminate Anthropic products by September 1st, leading to divided opinion within the agency. On the Chinese side, there have been reports of imitations of the Anthropic OpenAI model using fake accounts. Musk reversed his previous critical stance and praised Anthropic as an "AI leader," but at the same time required Tesla employees to use Grok, making the contradiction between his words and actions noticeable.

[OpenAI trends]

Apple's lawsuit dominated OpenAI coverage this week. The claim is that a former Apple employee stole trade secrets from the development of hardware for ChatGPT, and while the actual damage is still under dispute, the brand image of both companies has been severely damaged. In terms of products, the company has launched GPT-5.6 and "ChatGPT Work" for businesses, but has announced the discontinuation of the browser "Atlas" and is proceeding with business selection. In terms of human resources, safety managers continue to retire, and including the departure of Lilian Weng, this is the ``sixth safety manager change in two years.'' It was a week that highlighted the instability of the organization, with AGI head Fidji Simo also reportedly stepping down due to health reasons.

[Google AI/DeepMind trends]

Google released Gemini Pro 3.1, clarifying its strategy to offer lower prices than OpenAI and Anthropic. AlphaEvolve has moved to the official version (GA), and Gemini has entered the actual operation stage for use in automatically optimizing algorithms for scientific calculations, GPU design, and drug discovery. Gemini is being rolled out to low-priced devices such as Android Go, and an agent task function that runs in the background has been added. Gemini's roadmap has been reorganized in the wake of Google I/O, and Gemini is becoming more competitive with OpenAI and Anthropic in terms of both price and penetration rate.

[Trends of other major companies]

Meta introduced Muse Spark 1.1, a low-priced coding model, and shook up the market with a price point that was about one-fourth that of Anthropic and OpenAI. Zuckerberg has resumed posting X for the first time in three years due to this announcement. Meanwhile, Instagram's AI image generation feature was withdrawn following backlash over accusations that it was using public photos without users' consent. xAI introduced Grok 4.5, claimed to have achieved Opus-level performance at a low cost, and rose to second place in the coding benchmark APEX-SWE. However, Musk himself has contradicted himself by admitting that Grok is inferior to Claude, yet requiring Tesla employees to use it, which has led to lawsuits and complaints to the FTC pointing out that it is connected to crime. Microsoft has made GPT-5.6 a priority model in Copilot, but the adoption rate is low at less than 4.5%, and Microsoft is increasing its efforts to reduce the cost of dependence on OpenAI and Anthropic by investing in in-house AI. DeepSeek has cut prices by 75% and developed its own AI chips to reduce dependence on the United States, and authorities in Beijing are considering restricting overseas access to key Chinese models.

[AI celebrities' statements and trends]

Sam Altman said, "AI will create jobs across the board," in a shift from his earlier cautionary tone. The feud between Musk and Altman has reignited on X in the wake of the Apple vs. OpenAI lawsuit. Musk remains bullish, saying that SpaceX's corporate value could exceed the value of the Earth itself. Jensen Huang named Marvel as the next $1 trillion market capitalization stock. Anthropic co-founder Dario Amodei reportedly lost control to his co-founder at a White House meeting, and the departure of OpenAI's safety executives continues.

[Technological breakthrough]

This week was a week filled with changes to the base model generation. GPT-5.6, Claude Sonnet 5/Fable 5, Gemini Pro 3.1, Grok 4.5, and Muse Spark 1.1 were launched almost simultaneously, and competition is intensifying on both the axes of performance and price. AlphaEvolve's transition to GA indicates that the movement to incorporate generative AI into practical areas such as scientific calculations, GPU optimization, and drug discovery candidate search has moved from the experimental stage to the practical stage. The integration of browser tools into Claude Code is an example of the expansion of agents to directly interact with the external environment, continuing the shift from stand-alone chats to task execution tools.

[Regulation/Policy/Geopolitics]

The grace period under the EU AI Act will end on August 2, 2026, and from then on the European Commission will have the power to sanction the learning and behavior of frontier models. In the United States, the Trump administration is taking aim at state-level AI laws and regulations, and friction between the federal government and the states is expected. In the education field, the University of Chicago Law has banned AI devices from classes for first-year students, and the state of Illinois has banned the use of AI in teacher evaluations. In the copyright field, disputes over AI learning data continue in Australia, and Canada is considering an AI privacy bill called Bill C-36. A UK study has identified the potential for terrorism-related abuse as a common issue for major chatbots, and pressure is increasing on both safety and regulatory fronts.

[Change in power balance]

The structure of ``OpenAI being the strongest'' that has continued since last year has clearly been shaken during this period. While Apple's lawsuit and continuous departures from the safety department have damaged OpenAI's brand value, Anthropic is increasing its presence in both government procurement and corporate adoption. However, Anthropic's position within the government is still unclear, as evidenced by the contradictory moves of Air Force's exclusion policy and NSA's continued recruitment. Google and Meta are squeezing OpenAI/Anthropic's profit margins with their pricing strategies, and xAI's Grok 4.5 is making inroads into Anthropic's stronghold in the coding area. The Chinese are moving away from dependence on the United States through DeepSeek's price offensive and domestic chip development. Until a few months ago, the dichotomy of "OpenAI vs. Anthropic" was shifting to a multipolar one that includes Google, Meta, xAI, and DeepSeek, and this trend is expected to accelerate further within the year.

[Points to note in the next period]

We should pay close attention to the outcome of the initial trial of the Apple vs. OpenAI lawsuit and the impact it will have on OpenAI's corporate contracts and human resources recruitment. The focus will also be on the successor to OpenAI's head of safety and the succession system for Fidji Simo. With the end of the grace period for the EU AI Act (August 2nd), the test will be to what extent frontier model companies will take compliance measures in advance. In the price competition, the next focus will be on DeepSeek's additional price cuts, Meta Muse Spark's actual usage data, and Gemini Pro 3.1's corporate adoption rate, and how Air Force's Anthropic exclusion policy will be resolved in the run up to the September 1st deadline will also be a predictor of government procurement trends.