Growing concerns over artificial intelligence have prompted a technology specialist to highlight potential risks and explain what people living in the UAE should understand, including practical precautions and important considerations as AI continues to become part of everyday life.
AI Concerns Grow as Tech Expert Outlines What UAE Residents Need to Know.
Anthropic CEO Dario Amodei has called for greater caution in advancing frontier artificial intelligence, saying development could require a slower pace so that safety measures, regulatory frameworks and independent monitoring can evolve quickly enough to address potential risks.
AI Safety Concerns Are Increasingly Coming From Within the Industry
Warnings about the potential dangers of artificial intelligence are no longer limited to academics, campaigners or technology critics outside the industry. Some of the strongest calls for caution are now being voiced by executives, researchers and engineers who are directly involved in developing increasingly powerful AI systems.
Dario Amodei, chief executive of Anthropic, has suggested that progress in frontier artificial intelligence may eventually need to be slowed if safety measures and independent oversight cannot develop at a comparable pace. His comments reflect growing concern within the technology sector that the capabilities of advanced AI could progress faster than the systems designed to monitor and control them.
Other prominent technology leaders have also called for stronger safeguards. OpenAI CEO Sam Altman, DeepMind CEO Demis Hassabis and entrepreneur Elon Musk have each spoken publicly about the need to address the potential risks associated with increasingly capable AI. At the same time, researchers who have worked in artificial intelligence, including some who have since left major technology companies, have raised increasingly serious questions about the direction of the field.
The concerns cover a wide range of possible outcomes. Some researchers are examining scenarios in which advanced systems could eventually perform more tasks independently, require less direct human involvement or potentially contribute to the development of future AI systems. There are also discussions around whether highly capable models could become difficult for humans to understand, supervise or control.
Such possibilities can sound distant or highly theoretical. Yet not every AI-related threat belongs to the distant future. A number of risks are already becoming visible, particularly in cybersecurity, where criminals are beginning to use AI tools to make existing forms of digital crime more efficient.
For ordinary users and businesses, the term “AI risk” can therefore refer to very different things. It may describe immediate concerns such as phishing, identity theft and automated cyberattacks, or it may refer to much more speculative scenarios involving highly autonomous future systems.
That distinction is particularly relevant in the UAE. Artificial intelligence is being introduced across government services, companies and consumer applications at a rapid pace. As adoption increases, understanding the difference between current threats and longer-term possibilities becomes increasingly important.
Cybersecurity Shows How AI Risks Are Already Emerging
Cybersecurity provides one of the clearest examples of how artificial intelligence is changing the threat landscape today.
Yassin Watlal, Senior Director of Systems Engineering for the Middle East, Turkey and Africa at CrowdStrike, told Khaleej Times that AI is already making certain types of cybercrime more accessible to people with limited technical expertise.
According to Watlal, artificial intelligence has reduced some of the obstacles that previously prevented less experienced criminals from carrying out sophisticated digital attacks. Tasks that once required significant technical knowledge or considerable amounts of time can now be assisted or accelerated by AI-powered tools.
The development does not necessarily mean that cybercriminals are creating completely new methods of attack. Instead, AI can make familiar techniques considerably quicker, more scalable and more convincing.
One example is phishing. Attackers can use AI to create messages that appear more natural and contain fewer of the spelling, grammar and formatting mistakes that once made fraudulent emails easier to identify. Messages can also be adapted to particular organisations, regions or individuals, making them appear more relevant to their intended targets.
AI can also assist with impersonation. Voice technologies and other generative tools can make it easier for criminals to imitate people or create convincing communications designed to persuade victims to disclose information or authorise transactions.
Malicious software is another area of concern. AI-assisted development can potentially help attackers produce or modify code more rapidly, while automated systems can support other stages of an attack once access to a target has been obtained.
This acceleration is important because cyberattacks often involve multiple stages. An attacker may initially gain access through stolen credentials, a deceptive message or another weakness and then attempt to move deeper into an organisation.
Watlal pointed to an important change in how criminals approach corporate systems. Rather than attempting to defeat security controls through a direct technical intrusion, attackers may increasingly seek legitimate credentials belonging to someone who already has authorised access.
In other words, gaining access by appearing to be a legitimate user can sometimes be easier than attempting to break through the organisation’s defences.
“They will want to log in, not break in,” Watlal said.
The distinction highlights why cybersecurity is becoming increasingly connected to identity protection. Even organisations with substantial technical security measures can face problems if criminals obtain valid login details and use them to operate inside trusted systems.
Speed Has Become a Critical Factor
The growing speed of cyberattacks adds another layer to the problem.
CrowdStrike reported that the average period between an initial compromise and an attacker moving into other parts of an organisation dropped to 29 minutes in 2025. The company said its fastest observed example took only 27 seconds.
For cybersecurity teams, such figures illustrate how quickly an incident can develop once an attacker gains an initial foothold.
Traditionally, security professionals have relied on monitoring systems, alerts and human investigation to identify suspicious activity. Analysts may need to examine an alert, determine whether it represents a genuine threat, investigate the affected device or account and then decide what action should be taken.
When malicious activity can spread within minutes—or even seconds—the available response window becomes significantly smaller.
This is one of the practical consequences of AI-assisted cybercrime. The technology does not necessarily need to give attackers an entirely new form of attack to create a serious problem. Simply helping criminals perform existing activities faster can change the balance between attackers and defenders.
AI can potentially assist at several points in the attack chain. It can help generate convincing communications, analyse information about potential targets, automate repetitive tasks and support criminals as they attempt to move from one compromised account or system to another.
The result is an environment in which both sides are increasingly relying on automation. Security teams are using AI to identify suspicious behaviour, analyse large quantities of information and respond to threats, while attackers can use similar technologies to improve the efficiency of their operations.
That creates an ongoing technological race.
Separating Immediate Threats From Future Scenarios
The broader debate surrounding AI often combines risks that are already observable with possibilities that remain uncertain.
Cybercrime is an example of a current and measurable concern. Organisations are already dealing with phishing campaigns, stolen credentials, malicious software and other forms of digital attacks. AI can make some of these activities easier or faster.
By comparison, discussions about highly autonomous AI systems belong much further along the technological spectrum. Researchers are considering what could happen if future models become capable of carrying out increasingly complex tasks with limited human supervision.
Some experts have raised questions about systems that could potentially improve their own capabilities or contribute to the development of more advanced AI. Others are concerned about the possibility that future systems could behave in ways that are difficult for their creators to anticipate or control.
These scenarios remain the subject of research and debate rather than established outcomes. There is considerable uncertainty about how quickly such capabilities may emerge, whether they will emerge at all and what technical or regulatory measures could be developed in response.
That uncertainty is important when discussing AI safety.
Not every warning about artificial intelligence describes an immediate threat, and not every possible future scenario should be treated as an inevitable outcome. At the same time, the existence of uncertainty does not eliminate the need to consider how rapidly the technology is developing.
The challenge for policymakers, companies and members of the public is therefore twofold: responding to risks that are already occurring while preparing for capabilities that could create new challenges in the future.
Why the Issue Matters in the UAE
The discussion carries particular significance for the UAE because artificial intelligence is becoming increasingly integrated into the country’s economy and public services.
AI technologies are being used across a growing range of activities, from business operations and customer services to government applications and everyday digital tools. As these systems become more common, individuals and organisations are also becoming potential targets for AI-assisted attacks.
For businesses, this means cybersecurity cannot be treated solely as a technical issue. Employee accounts, passwords, access permissions and authentication systems can all become important points of vulnerability.
For individuals, the risks may appear in more familiar forms. A message that looks as though it comes from a colleague, bank, company executive or government organisation could potentially be generated or improved with AI. A voice call may also be harder to authenticate if criminals can convincingly imitate someone’s voice.
This makes basic digital caution increasingly important.
People should be careful when receiving unexpected requests for money, passwords, verification codes or sensitive information. Messages that create urgency or pressure the recipient to act immediately deserve particular scrutiny.
Organisations, meanwhile, need to consider how quickly they can identify and contain suspicious activity. Strong authentication, appropriate access controls, employee awareness and rapid incident-response procedures can all play a role in reducing exposure.
The broader lesson is that AI risk does not necessarily begin with futuristic machines operating independently. In many cases, it begins with existing criminal techniques becoming faster, cheaper or easier to execute.
A Changing Security Landscape
The rise of AI is therefore creating a complicated security environment.
The same technologies that can help criminals automate certain activities can also provide advantages to defenders. Security companies are developing AI-based systems to analyse large volumes of data, identify unusual behaviour and detect threats more quickly.
The central issue is increasingly about how quickly each side can adapt.
For organisations, a security strategy designed around slower-moving threats may not be sufficient when attackers can automate parts of their operations. Detection and response capabilities need to keep pace with the speed at which an incident can develop.
At the same time, the broader debate over frontier AI highlights another challenge. As technology companies develop systems with increasingly advanced capabilities, safety testing, oversight and governance need to develop alongside them.
The concerns expressed by technology executives and researchers therefore span two different time horizons. Some relate to problems that businesses and individuals are already facing. Others concern what could happen if future AI systems become substantially more capable and autonomous.
Keeping those categories separate can make the discussion easier to understand.
The immediate risks require practical cybersecurity measures and greater awareness. Longer-term questions require continued research, safety testing, independent oversight and informed public discussion.
For UAE residents, the key takeaway is that AI risk is not a single problem with a single solution. Some threats are already part of the digital environment, while others remain possibilities being studied by researchers.
As artificial intelligence becomes more deeply embedded in everyday life, understanding both sides of that debate will become increasingly important.
AI Is Changing the Fight Between Attackers and Defenders
Artificial intelligence is increasingly becoming a tool used by both sides of the cybersecurity battle.
Criminal groups can use AI to automate repetitive tasks, improve the quality of fraudulent messages and accelerate different stages of an attack. Security teams, meanwhile, are turning to similar technology to process large numbers of alerts, identify suspicious activity and respond to incidents more quickly.
Yassin Watlal, Senior Director of Systems Engineering for the Middle East, Turkey and Africa at CrowdStrike, said this growing dependence on AI means organisations have to consider both sides of the equation.
For defenders, speed is becoming increasingly important. A security team may receive thousands of alerts, but only a small proportion may represent genuine threats. AI can help analysts examine information more rapidly and identify patterns that might otherwise take considerably longer to uncover.
The same principle applies to attackers. If criminals can use AI to automate parts of their operations, the amount of time available for defenders to detect and stop an intrusion can become shorter.
This is one reason the discussion surrounding artificial intelligence cannot be limited to hypothetical future scenarios.
Much of the public conversation about AI safety focuses on the possibility of extremely powerful systems becoming difficult for humans to control. Those questions remain important, but they exist alongside risks that are already appearing in the real world.
Cybercrime is one example. AI can help criminals make convincing messages, imitate legitimate communications and carry out certain tasks at greater speed. These developments do not require the emergence of a hypothetical superintelligent machine. They are connected to technologies that are already available.
That distinction matters because it changes how individuals and organisations should think about AI security.
The question is no longer simply what artificial intelligence might be capable of many years from now. It is also about what current systems can already do and how quickly people can adapt their security practices to match those capabilities.
The Shift From Chatbots to AI Agents
There is another development that could make the conversation around AI safety more complicated: the growing emphasis on AI agents.
For many people, the first direct experience with generative AI was through a chatbot. A user would enter a question or instruction, the system would generate a response and the interaction would generally end there.
AI systems are increasingly being designed to do more.
Rather than simply answering a prompt, an agent can be assigned a broader objective and then carry out a sequence of tasks in an attempt to achieve it. Depending on how it has been designed, the system may be able to search for information, analyse documents, write or modify computer code, interact with software, access databases or use external digital tools.
This represents an important change in how AI systems operate.
A conventional chatbot might provide incorrect information. The user can then decide whether to accept, reject or verify that information before doing anything with it.
An agent can potentially move beyond providing information and take an action itself.
That difference introduces another layer of risk.
If an AI system misunderstands an instruction, the consequences may not be limited to an inaccurate sentence on a screen. Depending on its permissions, the system could potentially make changes to a file, send a message, interact with another application or initiate another process.
The technology is not automatically going to operate without human involvement. The level of autonomy depends heavily on how an organisation designs and deploys the system.
Nevertheless, the increasing capability of AI agents means companies need to think carefully about the boundaries placed around these systems.
Questions that once seemed theoretical are becoming operational issues.
What data should an AI system be permitted to access?
Which company applications can it interact with?
Should it be allowed to make changes without human approval?
What happens when an agent misunderstands an instruction?
How can an organisation detect an error before it creates a serious problem?
And perhaps most importantly, how quickly can a person intervene if an automated system begins doing something it was not supposed to do?
These questions are becoming increasingly relevant as businesses connect AI tools to internal systems and corporate data.
AI Needs Guardrails as Its Role Expands
Watlal said organisations should consider carefully what they allow AI systems to do and establish appropriate safeguards around those capabilities.
The idea is not necessarily to prevent businesses from using AI. Instead, it is about determining where automation is appropriate and where human involvement should remain mandatory.
An AI system may be highly effective at processing information, identifying patterns or completing repetitive tasks. But that does not automatically mean it should be given unrestricted access to sensitive systems.
Permissions can become particularly important when AI tools are connected to business databases, customer information, financial systems, software development environments or internal communications.
A useful approach for organisations is to establish clear boundaries before deploying an AI agent.
The company should know what information the system can access, which tools it can use and what actions it can perform independently. There should also be mechanisms for monitoring its activity and stopping it when necessary.
This becomes especially important when an AI system is capable of taking several actions in sequence.
The more steps an agent can complete without human intervention, the more important it becomes to understand what happens if one of those steps goes wrong.
A mistake at the beginning of a process could potentially influence everything that follows.
That does not mean autonomous AI systems are inherently unsafe. It means that their capabilities need to be matched with appropriate controls.
Human oversight remains particularly important when decisions involve sensitive information, financial consequences, security permissions or significant impacts on individuals.
The Longer-Term Question: Could AI Help Build Better AI?
Beyond the immediate concerns surrounding cybercrime and autonomous agents lies a much more difficult question.
Some researchers working on frontier AI are examining what might happen if artificial intelligence becomes capable of contributing substantially to the development of future AI systems.
AI is already being used by engineers and researchers in areas such as programming, data analysis, scientific research and software development. These applications can help people perform parts of the work required to build and improve AI models.
The long-term concern is not simply that AI will assist humans in those activities.
The question is what could happen if future systems become capable of performing a much larger portion of the process themselves.
For example, researchers have discussed hypothetical situations in which increasingly capable AI systems could assist with designing algorithms, improving software, analysing research or developing components needed for subsequent generations of AI.
If those capabilities eventually became highly advanced and increasingly autonomous, the pace of AI development could potentially change significantly.
This is the point at which discussions about recursive self-improvement and superintelligence enter the debate.
Recursive self-improvement generally refers to the hypothetical possibility that an AI system could help improve its own capabilities, with those improvements then enabling it to contribute to further improvements.
Whether such a process will actually occur remains uncertain.
There is no universal agreement among researchers about when highly advanced AI might emerge, what capabilities such systems would possess or whether they would behave in the extreme ways described in some risk scenarios.
Some researchers believe these possibilities deserve serious attention because the consequences of losing control over a highly capable system could be significant.
Others argue that many of the most extreme scenarios remain speculative and that excessive restrictions could interfere with technological progress that may provide substantial economic, scientific and social benefits.
The disagreement is therefore not simply about whether AI has risks.
There is already broad recognition that artificial intelligence creates challenges. The debate is also about the scale of those risks, how likely particular scenarios are and how society should respond when there is significant uncertainty.
Why Calls for Slower Development Are Being Discussed
When technology leaders and researchers call for greater caution or a slower approach to frontier AI, that does not necessarily mean they are demanding that existing AI tools disappear.
The proposals under discussion can involve a range of measures.
These may include more extensive testing before highly capable systems are released, independent assessments of safety, stronger security controls and greater transparency around the capabilities and limitations of advanced models.
Some experts also support allowing additional time between major capability developments so that researchers, regulators and companies can evaluate potential risks before moving to the next stage.
Another proposal involves greater cooperation among companies developing frontier AI systems.
Because the technology is advancing across multiple laboratories and countries, some researchers argue that safety cannot be addressed effectively by one organisation acting alone.
There are also calls for clearer rules concerning what advanced systems should be permitted to do, particularly when they are given access to sensitive information or external tools.
The details of these proposals remain contested.
There is no single agreed formula for balancing AI innovation with safety. Different researchers, companies and policymakers place different levels of emphasis on economic opportunity, technological progress, security and long-term risk.
The important point is that the debate is moving beyond a simple question of whether AI should be developed.
Increasingly, the discussion is about how advanced systems should be tested, deployed, monitored and controlled.
What Does This Mean for People in the UAE?
For ordinary residents, the response to these concerns does not require abandoning artificial intelligence.
The immediate risks are considerably more familiar than some of the long-term scenarios discussed by frontier AI researchers.
People are already encountering phishing attempts, fraudulent messages, impersonation scams and other forms of cybercrime. AI can make some of these techniques more convincing and potentially easier for criminals to carry out.
That means basic digital security practices remain important.
Watlal recommends using multi-factor authentication wherever it is available, particularly for important accounts. This provides an additional layer of protection if a password is compromised.
People should also be cautious when receiving unexpected requests involving payments, passwords, authentication codes or sensitive information.
A message that appears to come from a colleague, manager, financial institution or another trusted contact should not automatically be treated as genuine simply because it looks professional.
AI-generated content can make fraudulent communication more convincing.
If something seems unusual, verifying the request through a separate communication channel can reduce the risk of acting on a fraudulent message.
For example, instead of responding directly to a suspicious email requesting a payment, a person could independently contact the organisation or individual using a known telephone number or previously verified communication method.
Taking a few additional moments to verify an unusual request can be particularly valuable because modern scams are often designed to create urgency.
Attackers may try to convince victims that immediate action is required, leaving little time for careful consideration.
Slowing down can therefore become an important security measure in itself.
Businesses Face a More Complicated Challenge
The responsibilities are greater for organisations deploying AI internally.
Companies need to consider not only how criminals might use artificial intelligence against them, but also what could happen if their own AI systems are given excessive permissions.
An organisation introducing AI agents should establish clear limits around access to corporate information and digital systems.
Sensitive data should not automatically become available simply because an AI tool is capable of processing it.
Similarly, an automated system should not necessarily receive permission to perform every action that a human employee could perform.
A company might decide, for example, that an AI system can analyse documents but cannot send external communications without approval.
Another organisation might permit an agent to generate software code while requiring a human developer to review and approve changes before they are deployed.
The appropriate controls will depend on the specific use case.
The underlying principle is that AI capabilities and permissions should be considered together.
An increasingly capable system with unrestricted access can create a very different risk profile from the same system operating within a tightly controlled environment.
Humans Still Have a Central Role
Despite the growing capabilities of AI, human judgement remains an important part of the equation.
AI can process information quickly, identify patterns and perform repetitive activities at a scale that would be difficult for individuals to match.
But speed does not necessarily equal understanding.
A system may interpret a request differently from the way a person intended. It may rely on incomplete information or produce an apparently reasonable result that is nevertheless incorrect.
That is why human oversight remains important, particularly when an AI system is involved in decisions with significant consequences.
People need to understand what an automated system is permitted to do, what limitations it has and when its output should be independently checked.
This becomes even more important as AI systems move from simply generating content to interacting with real-world digital environments.
Two Different AI Questions
The debate surrounding artificial intelligence can ultimately be divided into two broad questions.
The first concerns the present.
How should individuals, businesses and governments protect themselves from criminals who are using increasingly capable AI tools?
That problem is already developing. Cybersecurity teams are responding to AI-assisted threats, while individuals are encountering increasingly convincing digital fraud.
The second question concerns the future.
How much independence should society give increasingly capable AI systems themselves?
That question is much more difficult because it depends on technological developments that have not yet fully occurred.
It involves decisions about autonomy, permissions, oversight, accountability and the limits that should be placed on advanced systems.
There is no universal agreement on how those questions should be answered.
Some technology leaders and researchers believe society should prepare early for potentially serious consequences. Others emphasise the uncertainty surrounding the most extreme predictions and the potential benefits of continuing rapid technological development.
But one aspect of the debate is becoming increasingly difficult to ignore: AI is no longer simply a tool that waits for humans to provide every instruction.
Systems are becoming capable of completing longer sequences of tasks, interacting with software and assisting with increasingly complex work.
That evolution makes There is no need to treat every discussion about advanced AI as a prediction of an imminent catastrophe. At the same time, current risks should developing advanced AI may disagree about exactly where the technology is heading and how society, businesses, researchers and policymakers a continuing one: understanding what AI can do today, preparing for what it may be able to do tomorrow and ensuring questions about control and accountability more immediate.
For UAE residents, the practical message is relatively straightforward. There is no need to treat every discussion about advanced AI as a prediction of an imminent catastrophe. At the same time, current risks should not be dismissed simply because some future scenarios remain uncertain.
Good digital security, careful handling of unexpected requests, strong authentication and awareness of AI-assisted scams can help address the threats people already face.
For businesses, the challenge is broader. Organisations need to consider both the threat posed by criminals using AI and the risks created by their own deployment of increasingly autonomous systems.
As AI becomes more deeply embedded in workplaces, government services and everyday applications, the distinction between a tool that simply provides information and a system capable of taking action will become increasingly important.
The people developing advanced AI may disagree about exactly where the technology is heading and how society should respond.
What appears increasingly clear, however, is that the conversation about AI safety is no longer only about distant possibilities.
Some of the challenges are already visible, while others may emerge as systems become more capable.
That makes the task for individuals, businesses, researchers and policymakers a continuing one: understanding what AI can do today, preparing for what it may be able to do tomorrow and ensuring that the systems being introduced into society remain subject to appropriate safeguards and human oversight.
