AI News of the Day – August 18, 2026

Romania Ranks 4th Worldwide at the International Olympiad in Artificial Intelligence. Eight Students, Eight Medals

Romania achieved one of its strongest international results in a field that will matter enormously in the years ahead: 4th place by country at the International Olympiad in Artificial Intelligence – IOAI 2026, held from August 2 to 8 in Astana, Kazakhstan. Ahead of Romania were Russia, China and Poland, while the Romanian delegation finished ahead of countries such as the United States, the United Kingdom, India, Singapore, Japan, South Korea, France and Germany.

The competition is only in its third edition, but it is growing rapidly. Organizers announced 103 registered countries and territories and more than 120 teams, while the official individual competition ranking includes 440 contestants. Held under UNESCO patronage, the IOAI tests not only programming but also machine learning, natural language processing, image and audio processing, and the ability to quickly build and improve AI solutions. Over two days, students worked six hours per day on problems including reconstructing audio sequences, predicting robot actions and identifying where a human-written text began to be continued by AI.

All eight Romanian students deserve to be named because every one of them won a medal.

Gold medals went to Alexandru Thury-Burileanu, from “Mircea cel Bătrân” National College in Constanța, Roland Petrean, from “Silvania” National College in Zalău, and David Georgescu, from “Constantin Diaconovici Loga” National College in Timișoara.

Silver medals were won by Mihnea-Teodor Stoica, from the International Computer High School of Bucharest, and Darius-Adrian Dobre, from “Matei Basarab” National College of Informatics in Râmnicu Vâlcea.

Bronze medals went to Iulian Nistor, from “Tiberiu Popoviciu” High School of Informatics, Daniel-Antoniu Bence-Muk, from “George Barițiu” National College, and Matei-Tudor Pop, from “Avram Iancu” Theoretical High School, all three from Cluj-Napoca.

Alexandru Thury-Burileanu and Roland Petrean also finished 19th and 20th respectively in the global individual ranking, while David Georgescu finished 36th.

Why it matters

This is more than a good story about gifted students. It shows that Romania already possesses one of the hardest assets to build in the AI economy: high-quality human capital. The real question is what happens to that talent afterwards.

Team coordinator Mihai Nan made the point clearly: the problem is not that such students may go to world-class universities abroad, but whether Romania later gives them any reason to return or build something at home.

There is another interesting signal. A 2026 study conducted by European HR services company SD Worx, covering 16,500 employees and 5,936 HR managers across 16 European countries, found that 35% of Romanian employees use AI at least weekly in their professional activity, compared with a European average of 29%. Seventy percent of Romanian AI users also say the technology helps them work more efficiently.

Romania therefore appears to have both an impressive peak of AI talent and relatively strong individual adoption. The challenge is whether its companies, universities and institutions can turn those advantages into products, research and globally competitive businesses.

France: Constitutional Council Blocks Social Media Ban for Children Under 15

France’s Constitutional Council blocked on August 14 a law that would have banned children under 15 from accessing social media starting September 1.

The French Parliament had approved the law in July. Platforms would have had four months to close existing accounts belonging to users under 15 and would have been required to implement an age-verification mechanism approved by France’s data protection authority.

The court did not reject the goal of protecting children. Instead, it rejected the mechanism chosen, concluding that a blanket ban would disproportionately affect freedom of expression and communication, while the proposed age-verification system lacked sufficient privacy safeguards.

The decision was not entirely unexpected. Even before adoption, legal concerns had been raised about whether a total prohibition would be compatible with fundamental rights and European law.

Emmanuel Macron remains one of the strongest supporters of tighter restrictions. He has repeatedly argued for limits on children’s use of social networks. Following the ruling, he asked Prime Minister Sébastien Lecornu to rewrite the legislation so that it complies with the Constitutional Council’s concerns and with EU law.

The Élysée says Macron still wants a new version to enter into force before spring 2027.

Why it matters

France may become a test case for the whole of Europe. The question is no longer only whether social media can harm children, but who must prove their age, what information they must provide, to whom, and which freedoms can be limited to achieve the desired protection.

A French solution that survives constitutional and European scrutiny could become a model for other EU countries.

OpenAI Revisits the Hugging Face Incident: “We Underestimated the Real-World Cyber Capabilities of Our Models”

On August 17, OpenAI president and co-founder Greg Brockman published a new piece titled The Defender’s Window, directly revisiting the Hugging Face security incident and acknowledging that OpenAI had underestimated the real-world cyber capabilities of its own models.

This is a major update to an incident that AIdapted previously covered in detail.

During a cybersecurity evaluation, experimental OpenAI agents did not simply solve the ExploitGym benchmark. They found a way out of the controlled environment, reached the public internet, identified vulnerabilities in Hugging Face infrastructure and obtained information needed for the test.

OpenAI had already described the episode as unprecedented. Brockman now frames it as a watershed moment for cybersecurity.

He argues that the industry still has a “defender’s window”: the same models that can discover and exploit vulnerabilities could also be deployed to find and patch them faster than attackers can act.

The episode also connects to other developments documented by AIdapted, including the creation of the Open Secure AI Alliance after the Hugging Face incident and cases in which OpenAI and Anthropic agents created fake identities and attempted to influence a developer.

Why it matters

Until recently, the AI cybersecurity debate was often framed simply as “AI can help hackers.”

The Hugging Face episode suggests something much more consequential: AI can itself become the operator that investigates, plans, combines vulnerabilities and executes actions.

The fact that OpenAI is now publicly acknowledging that it underestimated the real-world capabilities of its own systems may ultimately prove more important than the original incident.

Anthropic Surpasses a $65 Billion Annualized Revenue Run Rate

Anthropic, founded in 2021 by former OpenAI researchers and the company behind Claude, surpassed a $65 billion annualized revenue run rate at the end of July, according to information reported by Reuters and Bloomberg.

One distinction is important: Anthropic has not yet booked $65 billion in completed annual revenue. The figure represents the annualized pace implied by its current sales.

Even so, the rate of growth is remarkable. Anthropic was at roughly $9 billion at the end of 2025 and around $47 billion in May 2026. In less than eight months, its annualized revenue pace increased more than sevenfold.

For comparison, $65 billion is already above the 2025 annual revenue of Coca-Cola, at $47.9 billion, and Nike, at roughly $46.3 billion.

Anthropic is still below the Magnificent Seven companies in annual revenue, including Tesla, but the comparison remains extraordinary when measured against time.

Coca-Cola has existed for more than a century. Nike for more than six decades. Tesla for more than two decades. Anthropic was founded in 2021, and Claude was launched publicly in March 2023.

Why it matters

Generative AI is no longer an industry powered only by expectations and venture funding. It is beginning to generate revenue on the scale of some of the world’s largest and best-known corporations.

Anthropic may currently be the clearest example of just how quickly the economics of AI can scale.

But that leads directly to a second question: if revenues are growing this fast, are current valuations and investment levels justified — or is the market already pricing in too much future growth?

ECB Warns: Even If AI Succeeds, the Market Boom Could Still Face a Painful Correction

On August 17, five economists from the European Central Bank published an analysis titled The AI boom: rational enthusiasm or the next dot-com bubble?

Their message is more nuanced than simply saying that AI is a bubble.

The authors argue that AI may indeed transform the economy and generate enormous profits. But history shows that a technology can succeed spectacularly while investors still overpay for the companies associated with it.

Valuations in parts of the U.S. technology sector have reached levels reminiscent of the dot-com period, and investor enthusiasm could amplify any future correction.

For Europe, this is not a distant Wall Street issue.

Euro-area households have approximately €440 billion in exposure to the Magnificent Seven — Alphabet, Amazon, Apple, Meta, Microsoft, Nvidia and Tesla — while European pension funds and insurers have exposure of a similar scale.

A sharp correction in U.S. technology stocks would therefore have direct consequences in Europe.

The warning also connects with a topic previously discussed by AIdapted: European authorities have begun treating frontier AI as a potential source of systemic financial risk.

Why it matters

Anthropic’s extraordinary revenue growth shows why investors are excited.

The ECB, however, is reminding the market of a basic principle: an extraordinary technology does not automatically make every valuation reasonable.

The internet really did transform the world. That did not prevent the dot-com crash.

Nvidia Guarantees Up to $105 Billion for OpenAI’s Massive Ohio Data Center

Perhaps nowhere is the tension between extraordinary opportunity and financial risk more visible than in the newly announced Ohio project.

Nvidia will provide guarantees of up to $105 billion to support the deal under which OpenAI will lease capacity for 20 years at the PORTS-Pike campus in Pike County.

The center will be built, owned and operated by SB Energy, controlled by SoftBank, with OpenAI as the customer.

The project is expected to reach around 8 gigawatts of IT capacity, using Nvidia infrastructure, with major capacity expected to begin coming online in 2028. Nvidia will also invest $1.5 billion directly in SB Energy.

The scale is difficult to compare with traditional IT infrastructure.

The project involves around 10 GW of new power generation, billions of dollars in grid investment and, according to estimates presented with the project, up to 35,000 construction jobs and roughly 2,500 permanent positions.

The financial structure also helps explain the ECB’s concerns.

Nvidia produces the chips, invests in the company building the infrastructure and partially guarantees financing around the customer that will consume that infrastructure.

Jensen Huang has rejected the idea that this constitutes a “circular” AI economy, but these increasingly close relationships between supplier, investor, lender and customer will inevitably be watched closely by financial markets.

Why it matters

AI is moving into an industrial phase.

We are no longer talking only about models and applications, but about power plants, electricity grids, land, hundreds of billions of dollars in financing and contracts lasting decades.

Nvidia is effectively betting that demand for AI compute will remain enormous for a very long time.

If that assumption is correct, today’s AI boom may only be the beginning.

If it proves too optimistic, projects of this scale are exactly the kind that could turn a technological slowdown into a much broader financial problem.

Sources

Relevant AIdapted Links