Yesterday (Friday), the company “OpenAI” admitted that its artificial intelligence tools published on the Internet 53 images sent by users through its “ChatGPT” platform, in a new leak, this time extending to the data of the company’s own users.
The giant company in the field of artificial intelligence, which is headquartered in San Francisco, also revealed that its tools had penetrated websites belonging to American federal agencies, thus confirming information previously reported by the New York Times, but indicating that it had only collected information available to the public.
OpenAI explained that the links to the 53 images were not publicly available, and were mistakenly published on electronic hosting sites, and indicated that most of them were removed with the help of the relevant hosting service providers, and that work is underway to remove the remaining images, according to what Agence France-Presse reported.
These photos were sent from accounts whose owners agreed to use their data to develop the company's models. Before being used, it passed through a personal data privacy filter, and the company confirmed that the images could no longer be linked to the users who sent them.
OpenAI did not clarify whether these images showed identifiable people or sensitive data, and did not reveal further details in response to questions from Agence France-Presse.
The images were published as a result of the behavior of “artificial intelligence agents”, which is software designed based on artificial intelligence models and capable of performing a series of sequential tasks autonomously.
The company explained that proxy artificial intelligence tools being used in the framework of its research transferred training data to external services, most of which was not issued by users, and pointed out that most of the transfers took place before it enhanced security around its research in August.
It confirmed that it is currently conducting a comprehensive survey of the activities of its previous proxy tools, in a process that will take several months.
The company's CEO, Sam Altman, admitted on Friday that "we were not as fast as we had hoped" in the review and accident detection processes, justifying this by the huge volume of data that must be analysed.
An OpenAI spokesperson told Agence France-Presse that “the majority of the activities that have been subject to scrutiny so far have been standard search tasks, such as accessing publicly available content on the Internet to answer questions.”
He added, "Some of these activities included government websites, as our models often turn to them as reference sources for public information."
Thus, the company confirmed that its models accessed public information on the websites of the US Securities and Exchange Commission, the body that regulates financial markets, before publishing part of them on another page on the Internet, stressing that it did not find any trace of access to information that is not available to the public.
Qualcomm's journey into artificial intelligence data centers did not begin with one of the major cloud computing companies in the United States, but rather with a Saudi company. This was revealed by Cristiano Amon, CEO of Qualcomm, during a special interview with Asharq Al-Awsat on the sidelines of the “Snapdragon Summit 2026” conference in the American city of Hawaii, saying that Humain was the company’s first client when it began developing its vision and products to enter the data center market.
But the relationship that Amon describes does not stop at this project; Today, it extends to computers supported by artificial intelligence, industrial artificial intelligence with Aramco, local engineering and cars, while he sees future opportunities in robotics and other areas that the company has not traditionally entered in the same depth.
Amon stated that the company’s current projects in Saudi Arabia are “just the beginning,” pointing to “many intersections” between what Qualcomm is trying to build, and the Kingdom’s trend to adopt new technologies and invest early in trends that are reshaping computing markets.
Partnership with the Kingdom without borders
Amon wants the company's Saudi presence to go beyond the concept of a market in which it sells technology to a broader role in capacity building. After opening its engineering center in Riyadh and expanding its research and development presence, Qualcomm is making the Kingdom a center that serves the region, while searching for engineers in artificial intelligence, semiconductors, and modern computer platforms.
Amoun added that the scope of opportunities is not limited to current projects in data centers, computers, and the industrial sector, considering that robotics and cars are two additional fields, and that he sees “no limit” to partnership opportunities with the Kingdom.
These steps come within a broader cooperation network that includes “Humin,” “Aramco,” “KAUST,” and other Saudi partners, reflecting the transition of the company’s relationship with the Kingdom to more than one layer of the technology value chain. From research and engineering to infrastructure, devices and industrial applications.
From language to the physical world
One of the areas that Amon believes will lead the way next is what he calls “physical AI.” Unlike models that focus primarily on language, these systems deal with the physical world through sensor data, images, and signals from equipment and industrial environments, allowing them to be used in factories, energy, robotics, and vehicles.
Hence, Amoun links cooperation with Aramco to “artificial intelligence at the edge,” where data is analyzed close to where it is produced instead of always relying on a remote data center. He also refers to uses related to the Saudi Ministry of Interior, smart cities, and computer vision, in addition to cooperation with “Aramco” on fifth generation networks, especially for industrial environments.
He believes that the presence of local intelligence is particularly important in industrial operations spread across remote locations, where reliability and speed of response become part of the system requirements, and not just an improvement in performance.
He considered that industrial automation and robotics will be among the most prominent growth engines for “physical artificial intelligence,” which places the industrial sector alongside consumer markets among the growth areas targeted by the company.
The car is another computing platform
This trend intersects with the automotive sector, which has become an increasingly important part of Qualcomm's strategy. Amon pointed to “Sir”; The Saudi electric car company uses technologies from Qualcomm, as part of its vision to transform the car into a “connected computer on wheels.”
In-vehicle computing is no longer limited to entertainment or navigation, but extends to the digital cabin, control of vehicle functions, software-defined vehicles, and driver assistance and autonomy systems. He explained that Qualcomm is on its way to being the largest provider of semiconductor technologies to the automotive industry, pointing to its work with automobile companies across multiple categories.
He believes that AI agents are particularly suited to the car; The driver can interact with them vocally, while the systems make use of cameras, sensors and real-time context to understand what is happening in and around the vehicle. Thus, in Amon’s vision, the car becomes an extension of the same idea that brings together the phone, computer, glasses, and headphones: several different devices, but personal intelligence can move through them according to the context.
Why has the importance of the Middle East changed?
When I asked Asharq Al-Awsat why the Middle East has become a more important region for Qualcomm than it was several years ago, Amoun pointed to Saudi Arabia’s desire to transform into centers for technology, investment in sectors of the future, and diversification of economies. He stressed that this intersects with the path of Qualcomm itself, whose business was more focused on the phone market before it began expanding into new markets.
In Saudi Arabia specifically, Amoun linked the relationship to the Kingdom’s willingness to bet early on Qualcomm’s new direction. He once again cited the fact that “Humain” was the company’s first client in the data center project, considering that this resulted in a “bond of trust” and the desire to walk together on this journey. He added that this relationship was one of the reasons for the company's decision to increase resources and focus on the region.
When the phone starts working without you
At the Snapdragon Summit, the other part of the discussion moved from the markets to the device itself. Amon explains the shift to the “AI phone” with a simple idea: Previously, when a person was not using his phone, the device would wait. In the era of agents, the phone may continue to work even when the screen is not in front of the user, because an agent may perform a task, search, or interact with applications on its behalf.
This means that the phone is required to serve two parties at the same time. The user, and AI agents that the user has authorized to perform tasks. Amon says this imposes fundamental changes to the processor design and the amount of computing power, memory, and energy required.
Among what the company announced this year are processors with a maximum speed of 5 GHz, a speed that Amon said is achieved while maintaining a battery life that extends throughout the day, in addition to the ability of the highest category to run models exceeding 30 billion parameters on the device.
Qualcomm indicates that the Snapdragon 8 Elite Extreme Gen 6 supports Mixture of Experts models with more than 30 billion parameters, with 50 percent more shared memory. This year, the company also introduced two platforms in the premium category for the first time instead of just one option.
But Amon doesn't think the current borders will be enough for long; He spoke of the future need for dedicated artificial intelligence assistant units and more integrated computing and memory technologies, in light of the rapidly rising requirements for agents.
“Artificial Intelligence Phone”
Economically, Amon argues that these capabilities could create a new hardware upgrade cycle; Instead of buying a new phone just to get a better camera, longer battery, or higher performance, a new motivation may emerge when the AI capabilities available on new devices become different enough from previous generations.
He said that the industry is waiting for the transition from current smartphones to “artificial intelligence phones,” expecting this to create a demand for new categories and experiences. He believes that the rhythm of the phone market, in which Qualcomm is accustomed to developing a new generation every year, has given it the ability to deal with the current speed in the development of artificial intelligence, and to transfer this style of innovation to other markets.
Personal intelligence that is not tied to a single device
Amon goes further than the phone in defining the future of personal computing; When Asharq Al-Awsat asked him whether Qualcomm wanted Snapdragon to become the computing layer through which a single personal intelligence is transmitted between user devices, he replied that this was a “good way” to describe the trend. The company calls this “everywhere connected intelligence.” “Snapdragon” is present in phones, smart glasses, and the digital cabin inside the car, while the platform is expanding into new categories such as smart pendants, jewelry, and audio devices. Amon added that the concept of mobility in the future will not relate to a single device, but rather to a group of devices that will all become points of contact for artificial intelligence agents.
Audio devices
This trend also appears in audio devices; During the summit, Qualcomm unveiled the Snapdragon Sound Elite Gen 2 platform, which is designed to push headsets and wearable devices towards a role that goes beyond playing audio to acting as permanent interfaces for personal artificial intelligence.
The company says that the platform provides up to twice the artificial intelligence capabilities on the device compared to the previous generation, while consuming up to 40 percent less power, and supports direct connection to cloud services via Wi-Fi without always having to pass the connection through the phone. Targeted uses include smart assistants, translation, transcription and context understanding, in categories ranging from traditional headsets to audio glasses and new wearables.
The same idea appears in the advertisement for “Googlebook,” a new class of laptops running on the “Snapdragon X Elite” platform and designed around “Gemini Intelligence.” Qualcomm states that devices from Dell and HP will be among the first products in this category, with support for native Android applications, integration between computer and phone, and artificial intelligence capabilities on the device.
Equating trust and reducing friction
Personal intelligence that follows a user across multiple devices needs to know a lot about them, and this is where one of the most sensitive issues arises. Amon places the success of this experiment between two interconnected factors: They are trust and reducing friction.
If the agent does not know the user's context, it will not be helpful enough, and the person will find it easier to perform the task themselves. In contrast, the user will not give this context to an agent who does not trust how the data is accessed, stored, or used.
Amon states that the consumer must have a choice in determining which information can be sent to cloud services, and which information he wants to keep on the device, with a permissions system that determines what the agent can access. According to Amon, Qualcomm is working to build these mechanisms within the platform, so that the user can determine where data is stored and grant appropriate permissions. He stresses that this need for privacy and control, along with processing speed, are reasons why local on-device AI is important in parallel with the cloud. Amon did not present the matter as a solved problem, saying bluntly: “I don’t think we have all the answers.”
From personal memory to sovereignty
This leads to a broader concept of sovereign AI. Cristiano Amon sees sovereignty not only related to the way countries manage defense systems, financial services, or government functions, but it can also extend to citizens’ data and the extent to which they are able to control the parts they wish to make available to artificial intelligence agents. In his opinion, these questions provide greater opportunities for sovereign models that combine local capabilities in data centers with intelligence that works on devices, which is the context in which the company’s cooperation with “Human” also places.
When the agent acts on behalf of the user
The security problem becomes more complex when the agent not only knows the information, but also begins using the user's credentials and interacting with external services on the user's behalf. Amon argues that this calls for new levels of continuous authentication to ensure that the agent is actually working for the authorized person, with a greater role for biometrics and the storage of credentials and sensitive information within secure areas. This also includes what he described as the user's “personal memory map,” which in turn needs protection. He stresses that an important part of these guarantees must be supported at the hardware level itself, and not remain just a software layer added on top of the system, which he links to the security mechanisms present within the “Snapdragon” platforms.
The device and the cloud grow together
Despite Qualcomm's focus on local AI, Amon does not expect the device to replace the cloud. He gives the example of current mobile applications: the user often does not know, and does not need to know, which part of the application is running on the device and which part is running in the data center.
Artificial intelligence is expected to work in the same way; So that tasks move smoothly between the device and the cloud according to the required service, performance, privacy, and permissions specified by the user.
He added that he wants a future that gives the user clear authority over these choices, even if the load distribution process itself becomes transparent to him. Amon also rejects the assumption that increased computing on machines will reduce demand for data centers. The two “grow together,” he says; Some functions need to happen on-premises, while others require the greater capabilities of the cloud. As the number of agents increases and the frequency with which they are used throughout the day, the total demand for computing at both ends may increase rather than the load simply shifting from one place to another.
The computing ceiling has not yet been determined
This also explains Qualcomm’s decision to offer two platforms at the top of the “Snapdragon 8 Elite” category for the first time. The traditional flagship phone is still required to provide a better camera, higher performance, a longer battery, and advanced gaming experiences, but some proxy uses have begun to add a new level of demand for computing.
Amon stated that the industry is now going through a transition from “smart phones” to “artificial intelligence phones,” and it is not yet known where the upper limit of the capacity that new applications will need is. He added that artificial intelligence companies and users “want more computing,” so the company is still testing to what extent this demand can extend.
Betting success
Qualcomm's CEO does not measure the success of his bet on the agent era by processor speed, or by the number of models that can be run; When Asharq Al-Awsat asked him what he would accept as evidence of the success of the strategy in the next three years, he identified 3 indicators: The first: the transition of daily usage from an application-centric model to one in which people rely more on agents to accomplish tasks. The second is for the personal computing experience to be distributed across more devices, with the spread of glasses, watches, pendants, and other categories of personal intelligence. The third is the emergence of a clear upgrade cycle in which the consumer begins to say that he needs a new “artificial intelligence phone,” in the same way that previous technical transformations pushed users to new generations of devices.
But the success of this vision will ultimately depend on an as-yet-unsettled equation: the more an agent knows about the user, the more useful it becomes, and the more powers it is granted, the more sensitive questions about privacy, security, and control become. Between the desire to have artificial intelligence constantly working in the background, and the need to keep decisions and data under user control, Amon believes that the industry is still at the beginning of determining the limits that people will actually accept.
