As the two leading AI powers, the United States and China, form their own blocs, other countries are striving to capitalize on the strengths of both blocs and avoid losing out. Experts believe that the opportunity for countries to choose freely between these two superpowers in the field of artificial intelligence may shrink, which will force them at some point to choose between the “Pax Silica” bloc led by the United States and the “Global Cooperation Organization for Artificial Intelligence” led by China.
Kazakhstan...with the two superpowers
This may put many countries in a dilemma: Do they choose to use the latest American AI chips and technologies, or join the more cost-effective, open source Chinese AI systems?
Kazakhstan, a Central Asian country with a population of about 21 million, is a rare example of a country that has sought to bridge this gap. Just three weeks after Kazakh Minister of Artificial Intelligence Jaslan Madeev visited Washington last June to join the American group working to strengthen artificial intelligence supply chains, he headed to Shanghai to attend a signing ceremony that made Kazakhstan a founding member of China’s initiative to shape the global artificial intelligence agenda.
Joining the Pax Silica initiative means more technology partnerships, job opportunities in high-tech manufacturing, and a place in global artificial intelligence supply chains. Meanwhile, as part of China's Global AI Cooperation Organization, along with 28 other founding countries, including Pakistan and Russia, Kazakhstan also hopes to have a role in setting regional and global AI standards.
The Associated Press quoted Dana Masalimova, a former Kazakh diplomat, independent researcher and fellow at the Katherine W. and Shelby Collum Davis Center for Russian and Eurasian Studies at Harvard University: “Everyone is taking precautionary measures.”
Benefiting from American and Chinese investments
Kazakhstan hopes to benefit from the best of both blocs, and from American and Chinese investments... but it is in a “very critical situation,” according to Masalimova.
* Cooperation with America
Kazakhstan has rich biomineral resources necessary for the development of artificial intelligence, and cheap electricity to power data centers, which has become politically risky in the United States. Earlier this year, the Kazakh government signed an investment agreement worth up to $10 billion with Nvidia to create a “data center valley” in the country.
* Cooperation with China
At the same time, more Chinese companies are investing in Kazakhstan. Local officials say Chinese foreign direct investment in Kazakhstan will reach a record level of $2.8 billion in 2025. Chinese President Xi Jinping met Kazakh President Kassym-Jomart Tokayev last July to discuss ways to expand cooperation within the framework of China's Belt and Road Initiative to build global infrastructure.
Washington and Beijing are strengthening their technological influence
Both China and the United States are exploiting access to artificial intelligence to enhance their global influence. President Donald Trump insisted last month that “the winner is whoever wins in the field of artificial intelligence.” American officials and companies specializing in artificial intelligence have accused China of “extracting” capabilities on a large scale from leading American models. However, Beijing rejected these allegations, asserting that the United States seeks to monopolize artificial intelligence, while trying to restrict China's access to some of the most advanced artificial intelligence chips and manufacturing machines in the world.
“Pax Silica” implicitly targets China
Washington indicated that the “Pax Celica”, which implicitly targets China, is intended only for those it trusts. “States cannot be trusted as partners in Pax Celica when they actually engage in initiatives that promote a competing vision of superintelligence,” the State Department said, using the new term favored by the Trump administration for artificial intelligence.
Inclusiveness of the Organization for the Cooperation in the Field of Artificial Intelligence
For its part, Beijing praises the openness and inclusiveness of its Global Organization for Cooperation in the Field of Artificial Intelligence. The Chinese Embassy in Washington said in a statement: “As two leading countries in the field of artificial intelligence, China and the United States should benefit from their strengths in this field instead of competing with each other. Artificial intelligence should not be limited to major countries.”
Kyle Chan, a researcher at the John L. Thornton, of the Brookings Institution, noted that the messages adopted by the two major powers in the field of artificial intelligence are different, and this is mostly due to Beijing’s need to win over third countries in its effort to build an alternative coalition, while Washington already has a strong network of allies.
Indonesia and Singapore are in a critical position
This situation reinforces the power of what Chan described as “swing states,” as Indonesia and Singapore may also find themselves caught between China and the United States.
Like Kazakhstan, Indonesia has vital resources for artificial intelligence and other technologies, such as nickel and copper. Indonesia is also seeking to make the most of the two powers of artificial intelligence, after joining the Chinese Global AI Cooperation Organization.
Indonesia wants to benefit from both, Mari Pangestu, vice president of the Indonesian National Economic Council, told reporters at an event organized by the University of Hong Kong Business School. She added: “We should be able to participate in both, and choose the most appropriate system.”
American AI companies are investing in other countries such as Singapore, where Anthropic is setting up its latest office in Asia. The city-state says it is also considering joining the China Artificial Intelligence Organization.
Saudi Arabia... cooperation with America and China
In the Kingdom of Saudi Arabia, Humain, a startup company in the field of artificial intelligence, unveiled in September a platform built on an open source model from a Chinese company specializing in artificial intelligence. The Kingdom also concluded an agreement with NVIDIA to expand the scope of artificial intelligence-based computing using its advanced chips.
France...a private road
France, one of the most ambitious countries in developing artificial intelligence after the United States and China, is seeking to pave its own way through its emerging company, Mistral, whose new model launched this month outperforms some advanced Chinese models.
Japan is with America... and with non-dangerous Chinese applications
Japan is closely linked to the United States on strategic and security issues, including advanced semiconductors, reliable artificial intelligence, and standards. But it will not reject cheaper Chinese AI models for “lower-risk applications,” said Takahiro Tsuchiya, a professor at Kyoto Foreign Studies University.
Cyber threats facing organizations are no longer limited to external attacks, but rather risks emerging from within business environments pose an increasing challenge to information security and business continuity.
A serious fraud incident in an Italian bank
In an incident that reflects the extreme seriousness of this transformation, judicial authorities in Milan revealed a month ago that a prominent Italian bank had been subjected to a massive cyber fraud that resulted in a loss of $108 million. In the hack, the attackers relied on artificial intelligence tools to infiltrate digital messaging channels, make phone calls with a fake voice, and present themselves as well-known executives at the bank and the law firm that the bank works with, causing huge financial transfers amounting to 93 million euros under the guise of false reliability.
This incident shows that AI-powered threats are no longer limited to simple phishing, but rather include exploiting trusted privileges and simulating behavioral patterns of officials within closed environments. With the rapid development of these tools, organizations are facing an unprecedented challenge regarding how to distinguish between official transactions and those managed by malicious AI tools and agents.
A study on the “new intruder”
In this context, Trellix, a company specializing in the field of cybersecurity and protection solutions supported by artificial intelligence, revealed a report entitled “Mind of the CISO: The New Insider,” which highlights the impact of artificial intelligence and independent systems in shaping internal risks and the consequences resulting from the misuse of trusted access permissions and the extent to which this reflects on the role of information security chiefs.
Study methodology
The company commissioned an independent research agency to conduct a survey of 600 CIOs or senior security officials across the Middle East, the Americas, Europe, and the Asia-Pacific region, during the months of August and September 2026. The Middle East sample included 40 participants from the Kingdom of Saudi Arabia and 40 participants from the United Arab Emirates. Average costs were calculated based on responses from participants whose organizations had experienced internal security incidents in the previous 12 months and who provided data on the associated costs.
As for the results referred to at the Middle East level, data is collected from participants from Saudi Arabia and the Emirates.
Costs of security incidents: financial losses, operational and reputational impacts
The results of the study showed that the average cost of security incidents related to internal threats in organizations in the Kingdom of Saudi Arabia amounted to $1.95 million during the past twelve months. Although this amount is large, it represents only one aspect of the damages. The study showed that 63 percent of the institutions that witnessed internal incidents suffered the loss or leak of sensitive data, while 43 percent of them suffered from the disruption of operational processes, and 45 percent reported damage to their reputation and loss of customer confidence. These numbers confirm that a single security incident can threaten data, disrupt services, and harm an organization's main business relationships.
Threat drivers: from human negligence to “machine speed”
The reasons behind this damage vary, as unintentional errors or negligence by employees and trusted users accounted for 47 percent of internal incidents, while intentional attacks by humans caused 31 percent of them. What is striking is that operations carried out by artificial intelligence and autonomous systems alone account for 20 percent (one in every five incidents) of internal security failures. As reliance on these systems expands and they operate with reliable authority and at high speed (machine speed), errors or misuse may spread before security teams can intervene. The company warns that the absence of strict governance and rapid response mechanisms will exacerbate the financial losses of institutions in the future.
The company's head of information security confirms that with artificial intelligence agents working within the trusted environments of organizations and at breakneck speeds, the most resilient organizations will be those that subject these systems to governance as identities with high-level access powers, in parallel with the development of self-driving security operations centers (Agentic SOC) supported by artificial intelligence agents, capable of responding to threats at the same speed at which they develop, instead of relying on traditional human work procedures.
Pressures on CISOs and a gap in board awareness
The growing role of autonomous systems adds a new burden to CISOs on top of traditional human risks. In Saudi Arabia, 100 percent of CISOs surveyed agreed that their responsibility for managing internal risks had increased over the past year, while 70 percent of them expressed concern about the consequences and personal liability in the event of an internal security incident. In contrast, 80 percent of them indicated that senior management and board members lack a clear understanding of the nature of internal risks and their potential impacts, which puts CISOs under pressure to explain a wide range of trusted activities and their business implications to departments that may not yet understand the entire landscape.
Smart confrontation path
On the other hand, the results highlight the opportunities available to information security chiefs to enhance their security system, as 80 percent of cybersecurity officials in the Kingdom acknowledged that there is no clear vision of how employees use artificial intelligence tools, and 80 percent reported that their organizations are not fully prepared to manage the risks associated with these independent tools and systems. Improving visibility and ownership of “non-human identities” will give security leaders a clearer basis for identifying risks and making timely decisions.
But the comprehensive vision is not fully valuable unless it is coupled with rapid response, as it takes organizations in the Kingdom an average of 5.1 hours to discover an internal incident, and another 8.4 hours to respond to it, which is equivalent to 13.5 hours in total. This time gap underscores the urgent need to investigate suspicious activity in context and contain it quickly, especially when autonomous systems act at breakneck speeds.
The company confirmed that it is seeing a growing interest in understanding how reliable artificial intelligence systems work and how to deal with threats before they turn into losses that affect the business. Today's cybersecurity teams need to leverage their security data and respond more quickly. For CISOs, this requires the ability to track the activities of users and AI agents, analyze them in context, and contain any malicious activity before its impact spreads.
The rise in AI-powered attacks can be responded to using AI, through the Trellix Wise for Private Cloud solution, which is an AI-powered core engine for autonomous security operations centers deployed within the enterprise environment. This solution is designed to reduce the gap between detecting and responding to threats through faster investigations, while keeping security data under the organization's full control.
Researchers at the Georgia Institute of Technology have developed a digital forensic tool capable of recovering parts of deleted conversations from assistants running locally on Android phones, and then reusing the recovered information to build a new interactive session. Experts promise that deleting a conversation with an artificial intelligence assistant on the phone may not remove all traces of it from the device.
The tool, called ORISA, targets assistants that run large language models directly on the phone, rather than relying entirely on external cloud services. In these cases, parts of the conversation can remain in the device's working memory for some time even after they are deleted from the application interface.
Recover what remains in memory
ORISA relies on memory analysis techniques to search for parts of data that were not actually erased after the session ended or was deleted. The researchers point out that deleting a conversation does not necessarily guarantee that every temporary copy of its content will be removed immediately, especially if some data is still present in the operating memory.
In one test, the team created a virtual assistant called “Project Note Keeper” using the “ChatterUI” application on Android, then provided it with 100 fake information about a project before deleting the conversation.
When the researchers later tried to restore the session using ORISA, the reconstructed version was able to retrieve 26 pieces of that data. In contrast, the original session before it was deleted was able to access only 14 pieces of information, because the rest of the information fell out of the model’s active memory. This result shows that some information may remain in the device's memory even after the model itself stops actively using it during a conversation.
Tests across different applications and models
The researchers tested the tool on ten Android applications and three artificial intelligence models, in 60 different test settings. In these tests, the tool was able to recover information from conversations when parts of them remained in the phone's memory. The results do not mean that every deleted conversation can be completely restored. Researchers point out that the retrieval process becomes more difficult the longer the phone is used, because new activities may overwrite old data in memory. The study also focuses on local assistants working on Android devices, and does not provide evidence that conversations deleted from cloud AI services can be recovered in the same way.
Valuable for digital investigations
The researchers believe the tool could give digital investigators an additional way to see what type of information a user has shared with an AI assistant, as well as the nature of the help or answers they received from it.
Ken Xu, the study's lead researcher, said this technology could help determine what information a user provided to the assistant and what they received in return, even in cases where the conversation is no longer visible within the app. However, the ability itself raises privacy questions, especially with the increasing use of artificial intelligence models that work directly on phones to keep data away from servers.
