1. Google drops visible watermark for AI-generated content

Google has announced that users will be able to remove the visible watermark from images, videos, and audio materials generated with its AI tools. The option is available for creations made with models such as Nano Banana and Omni, and it can be disabled in the settings of Gemini and the Flow video editor; for Google Search, the feature will become available later. Importantly, removing the visible mark does not remove invisible identification mechanisms: Google retains SynthID and C2PA metadata so that content can still be identified as AI-generated.

The decision is important because it shifts the discussion about the authenticity of AI content from “can you see the watermark?” to “how can we technically prove a file’s provenance?” At a time when AI-generated images and videos are becoming increasingly difficult to distinguish from real ones, the visible watermark was a simple but imperfect solution. Google is now choosing a combination of creator freedom and invisible technical identification, a direction that becomes even more relevant following the implementation of the new transparency obligations under the European AI Act.

Why it matters: AI content may become harder to recognize with the naked eye, increasing the importance of metadata and digital provenance standards.

Sources: TechCrunch – Google will now allow users to remove visible watermark from its AI generations; The Verge – You can now turn off Google Gemini's visible watermarks.

2. OpenAI and Anthropic enter price war with Chinese rivals

OpenAI and Anthropic are sharply cutting the prices of some mid-tier AI models in response to pressure from Chinese developers such as DeepSeek and Moonshot. The Financial Times reports price reductions of up to 80% for certain OpenAI services and around 50% for some Anthropic services, amid customer demands for lower costs and the narrowing performance gap between Chinese and American models.

The real issue is not simply how much a prompt costs. The market is shifting toward the cost of completing an AI task, especially as AI agents execute complex processes and consume large amounts of computing power. If Chinese models can deliver comparable performance at significantly lower prices, American companies will have to demonstrate that their technological advantage justifies the difference. This is one of the first stages in which AI competition is beginning to resemble a mature market, where economic efficiency may matter just as much as raw performance.

Why it matters: the cost of AI could become one of the main factors determining which models are adopted at global scale.

Source: Financial Times – OpenAI and Anthropic in price war as Chinese AI rivals gain ground.

3. AI agents used in near-autonomous cyberattack against Taiwan’s government

In July, hackers suspected of having links to China used open-source AI agents in an operation targeting government systems in Taiwan. According to Israeli cybersecurity company Dream, the attack lasted four days, with up to eight AI agents working simultaneously to identify vulnerable systems, obtain credentials, and adapt their strategy based on the results. Twenty-one government systems were mapped, at least 85 accounts were compromised, and more than 2,500 personnel records were extracted. The operation even reached the systems of a Taiwanese nuclear safety agency and companies in the energy sector.

Taiwan separately confirmed that in July it detected an “abnormal” attack against several government agencies originating from outside the country, in which traditional techniques were combined with AI agents, including OpenClaw. Authorities said they detected the attack and managed to contain it. There is no official confirmation, however, that all the elements described by Dream and the Financial Times refer to exactly the same operation, and attribution of the attack to China remains an assessment rather than an official confirmation.

Why it matters: if Dream’s information is accurate, this is one of the clearest demonstrations yet that AI agents can move beyond simply automating individual hacking steps to autonomously coordinating a complex operation, including reconnaissance, exploitation, credential theft, and tactical adaptation. The issue therefore becomes strategic: a human attacker can deploy multiple agents simultaneously, operating at machine speed and drastically reducing the time available to defenders for detection and response. The Financial Times describes the case as a possible new stage in cyberwarfare.

Relevant Aidapted connection: the subject should be viewed in the context of the Aidapted articles “AI agents from OpenAI and Anthropic created fake identities and tried to persuade a developer to accept malicious code” and “2026 assessment: To get a good grade, they stole the solutions and grading rubric during the exam”, which highlight another dimension of the risk: AI agents capable of using social engineering and manipulating people to achieve their objectives.

Sources: Financial Times; Reuters; Taiwan Ministry of Digital Affairs / Focus Taiwan.

4. Nvidia seeks to turn AI infrastructure into a new asset class

Nvidia announced on August 10 a partnership with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to create financing platforms capable of mobilizing more than $500 billion in third-party capital for AI infrastructure. Nvidia could guarantee up to $125 billion of these financings. The initiative aims to enable developers, companies, governments, and cloud providers to build AI infrastructure without having to bear the entire cost directly on their balance sheets.

The economic significance is enormous: AI is beginning to be treated not just as software, but as long-term financeable infrastructure, similar to other industrial assets. At the same time, legitimate questions about risk are emerging: if demand for data centers and GPUs slows, some of these investments could become difficult to justify economically.

Why it matters: the AI race is entering a new phase in which the question is no longer just “who has the best models?” but also “who can finance the trillions needed to build the infrastructure?”

Sources: Reuters – Nvidia partners with Wall Street giants to raise $500 billion for AI buildout; Reuters – Goldman in talks with investors on Nvidia financing deal.

5. Microsoft reorganizes Copilot and drops some AI features

Microsoft is consolidating its Copilot apps for consumers and professionals into a unified experience, as part of a broader strategy to transform Copilot into a much larger AI platform. The rollout begins in August on mobile and web, with Windows and Mac versions to be updated later. At the same time, Microsoft is dropping several experimental features, including Podcasts, Deep Research, and Group Chats.

The move shows that the industry is beginning to enter a more pragmatic phase: not every AI feature that can be built needs to be kept. Microsoft is trying to simplify the product and bring users into a single Copilot ecosystem rather than maintaining a collection of separate experiments. It is a sign that, after the explosion of AI features over the past two years, natural selection is beginning to take place among AI products.

Why it matters: the AI market is moving from the “add AI everywhere” phase to one in which companies have to demonstrate that AI features are genuinely useful.

Source: The Verge – Microsoft is combining its Copilot apps ahead of a 'super app'.

6. France: Constitutional Council blocks social media ban for children under 15

France’s Constitutional Council has blocked legislation that would have banned children under 15 from accessing social media, ruling that the provisions infringe on freedom of expression and communication and do not provide sufficient safeguards regarding the age-verification mechanism. The law had been passed by Parliament in July and would also have required the closure of existing accounts belonging to minors under 15.

The decision does not mean France is abandoning the idea. Emmanuel Macron has already called for a revised version to be prepared, while the debate over protecting children on social platforms continues in France and at the European level. The issue is becoming one of the most complicated intersections between child safety, age verification, privacy, and freedom of expression.

Why it matters: the French approach could influence future European rules governing minors’ access to social media and age verification.

Sources: Reuters – France's top court blocks social media ban for under-15s; Le Monde – French government's social media ban suffers constitutional setback.

7. Europe is building its own AI infrastructure: Mistral targets 1 GW of computing capacity by 2030

Mistral AI is expanding its strategy beyond model development and is pursuing the construction of large-scale European computing infrastructure. The French company is targeting 1 GW of computing capacity by 2030, with a strategy that includes European data centers and partnerships with companies such as ASML. The goal is to enable Europe to control a larger share of the AI value chain, from infrastructure and models to applications.

Mistral has also entered into a partnership with European companies such as ASML. It is one of the most important examples of Europe’s strategy of linking AI models to industrial infrastructure. ASML invested €1.3 billion in Mistral in a funding round that valued the French company at nearly €12 billion, while Mistral is expanding its infrastructure and seeking to integrate across the entire AI value chain.

This strategy is directly linked to the issue of European digital sovereignty. Europe has strong companies in semiconductors, industry, energy, and infrastructure, but it does not yet have a European equivalent of the Nvidia–Microsoft–Google–Amazon combination. Mistral is attempting to occupy part of this space, and its own computing infrastructure is becoming essential to that ambition.

Aidapted has already analyzed this transformation in the article “Mistral wants 1 GW of European computing capacity by 2030. European digital sovereignty turned into a commercial product”.

Why it matters: European AI sovereignty is increasingly being measured in GPUs, data centers, and computing capacity, not just legislation. The European AI race is evolving from a quest for “a European chatbot” into an attempt to build a complete ecosystem.

Sources: Le Monde – Mistral AI aims for full value-chain presence; Financial Times – ASML and Mistral European AI deal.

8. AI is the best-funded startup sector in Europe in 2026

Artificial intelligence has become the best-funded startup sector in Europe. Tech.eu data shows that European AI companies raised €7.8 billion in 2026 through July 26, across 179 funding rounds, meaning AI accounted for 14.1% of the sectoral funding tracked. In the first half of the year alone, European AI companies had already raised €6 billion, 177.3% more than in H1 2025.

The data also shows a high degree of concentration: the ten largest funding rounds accounted for more than 80% of the capital raised by the sector. In other words, Europe is generating increasingly large amounts of AI capital, but a significant share is going to a relatively small number of companies. This helps explain why infrastructure projects and European companies capable of operating at scale are becoming so important.

Why it matters: Europe can no longer simply be described as “behind” in AI; the question is whether it can turn massive investment into globally competitive technology leaders.

Source: Tech.eu Funding Explorer – AI funding in 2026.

9. Bucharest will become a meeting point for AI, cybersecurity, and the digital economy in September

On September 29–30, 2026, Bucharest will host Impact SEE, an event featuring dedicated tracks on AI, cybersecurity & digital resilience, the digital economy, fintech, and startups. Speakers include Ryan Reynolds, Anna Lembke, and Diella, presented by the organizers as Albania’s Minister of State for Artificial Intelligence. The event is scheduled to take place at Romexpo/Face Convention Center and is expected to bring together around 200 speakers and 3,000 international participants.

For Romania’s AI ecosystem, the significance goes beyond visibility. The presence of AI, cybersecurity, investment, and digital transformation on the same agenda shows that AI is increasingly being treated as economic and geopolitical infrastructure, rather than simply as a technology. The case of Diella is particularly interesting for Romania because it represents one of Europe’s most unusual experiments involving the use of AI in public administration.

Why it matters: Bucharest is seeking to position itself as a regional meeting point for technology, capital, government, and industry.

Source: Impact SEE – Official website.