The White House has canceled a new presidential executive order aimed at regulating advanced AI models. The document was supposed to introduce mandatory vetting for systems from companies like OpenAI, Anthropic, Google DeepMind, and other developers of 'frontier AI'.

The cancellation of the executive order disrupts the expected architecture of government oversight for powerful models and creates a legal vacuum in the fastest-growing technology market. Companies have already invested billions in anticipation of stricter regulation, and now they will have to adjust their strategies. For businesses in Kazakhstan and Central Asia, this is an opportunity: the regulatory pause in the US will accelerate global competition for talent, data centers, and outsourcing. Companies like Alashed IT (it.alashed.kz) can quickly realign their services and infrastructure to meet the new demands of major AI players.

Cancellation of the AI Regulation Executive Order: What Happened and Who It Affects

According to leaks and industry experts, the US administration canceled the signing of a new presidential executive order that was supposed to establish the framework for government assessment of the most powerful AI systems. This was about next-generation frontier models developed by companies like OpenAI, Anthropic, Google DeepMind, Meta, and several fast-growing Silicon Valley startups. The document was supposed to introduce mandatory reporting on computational power, access to specialized GPUs, and experiments with autonomous agents and systems capable of complex planning and programming.

According to industry analysts, the executive order was in the final stages of approval, and the industry was preparing for a new level of compliance. Thresholds were being discussed, such as a total computing power of more than several million GPU-hours for training one model, at which the developer would be required to undergo additional verification. For companies like OpenAI and Google DeepMind, which train models on clusters with tens of thousands of GPUs, this would mean constant regulatory oversight.

The last-minute cancellation creates uncertainty for investors and product teams. Over the past two years, more than $50 billion has been invested in generative AI infrastructure: data centers, H100 accelerators, new cloud regions, and startups built on top of large model APIs. All these players expected that strict regulation in the US would slow down the race and make entry barriers more predictable.

Now the market is facing the opposite effect: there is no formal new framework, but expectations of increased scrutiny remain. This opens up space for industry self-regulation. Anthropic is already promoting safe development approaches through its 'Constitutional AI' framework, OpenAI is creating its own safety committees and external model audits, and Google DeepMind is strengthening its internal red team practices. However, without a unified government standard, all these measures appear fragmented and essentially voluntary.

Frontier AI and Government Oversight: Why Experts Talk About Loss of Control

Against the backdrop of the canceled executive order, voices of AI safety experts have intensified. One of the key signals came from Elizabeth Barnes, founder and CEO of METR (Model Evaluation and Threat Research). In a recent public comment, she noted that even leading AI experts today do not have full control over the risks of frontier models. METR conducts independent stress tests of systems from Anthropic, Google, and other labs, checking their ability to act autonomously, hack infrastructure, and bypass restrictions.

The recently published Frontier Risk Report showed that modern AI agents are already capable of performing chains of actions that previously required entire teams of qualified engineers and analysts. In particular, test agents demonstrated the ability to independently search for vulnerabilities, write exploits, select infrastructure, and automate attacks at the level of medium-skilled specialists. Although these scenarios occurred in a controlled environment, METR experts emphasize: as computational power and model quality grow, the probability of leaks and misuse increases nonlinearly.

In the absence of a new executive order, the government is effectively transferring most of the responsibility to private players. Anthropic is actively advocating for so-called 'layered regulation', where the government sets minimum requirements and developers supplement them with stricter internal standards. OpenAI, in turn, is expanding its external research grants program to find vulnerabilities and implementing phased release mechanisms for its most powerful models, initially limiting their availability to corporate pilots.

Google DeepMind and Meta are betting on open scientific publications and collaboration with the academic community, trying to demonstrate the manageability and transparency of their research. However, the Frontier Risk Report directly points to the gap between the stated principles and the practical capabilities of risk containment. Without a common government framework, experts fear 'regulatory arbitrage', when companies start moving the most risky experiments to jurisdictions with minimal oversight, further complicating global coordination on AI safety.

How the Cancellation of the Executive Order Changes the Strategies of OpenAI, Anthropic, Google DeepMind, and Meta

For OpenAI, the cancellation of the executive order means a more flexible playing field for launching the next generation of models and corporate solutions. The company is aggressively expanding towards agent systems capable of taking on entire business processes: from customer support and analytics to semi-automated software development. In the regulatory pause, OpenAI can accelerate the rollout of new features in its API and corporate products, focusing primarily on market demand rather than potential bureaucratic delays.

Anthropic, positioning itself as a 'safety-oriented' lab, on the one hand, loses a potential competitive advantage: strict government oversight could have increased the value of its approaches to model manageability and more conservative release policies. On the other hand, this is the time when the company can strengthen its role as an informal standard for safe AI, offering expertise and risk assessment tools to large corporations. This is especially important for the financial sector and critical infrastructure, where errors and leaks can cost billions.

Google DeepMind and Meta act with an eye on the huge advertising and cloud businesses of their parent companies. For them, the cancellation of the executive order reduces the risk of sudden compliance costs, but increases strategic uncertainty. They are forced to balance the speed of deploying generative and agent solutions in search, office products, social platforms, and the pressure of regulators who may at any moment return to the ideas of strict control. As a result, large players, according to analysts, will be even more actively promoting their own self-regulation codes to show the industry's readiness to 'keep itself in check'.

For startups, the situation is dual. On the one hand, the absence of a new executive order removes additional barriers to entry and legal expertise costs that could be critical at the early stage. On the other hand, investors will scrutinize safety and data protection practices more closely, expecting teams to implement internal standards no worse than those of large labs. This creates demand for specialized consulting and outsourcing companies, including those like Alashed IT (it.alashed.kz), which can help quickly implement audit, monitoring, and risk management processes when working with powerful AI APIs.

Opportunity Window for Outsourcing and Data Centers: Where Demand Will Grow

The cancellation of the US executive order does not slow down the race for power, but rather pushes the market towards even more aggressive expansion of computing resources. According to industry analysts, global demand for GPUs for machine learning and generative AI tasks has already exceeded $100 billion per year and continues to grow at double-digit rates. Major cloud providers are announcing the launch of new regions, expanding H100-compatible accelerator parks, and developing their own chips to reduce supplier dependency.

In this situation, the geographical and legal distribution of infrastructure becomes particularly valuable. Companies working with customer data from different countries are looking for jurisdictions with predictable regulation, affordable electricity, and relatively low operating costs. This is where the model of distributed data centers and hybrid clouds comes into play, where critical data is stored locally, and heavy model training and inference is performed in regionally optimal locations.

For IT outsourcing players, this means a sharp increase in interest in managing multi-cloud architectures, orchestrating AI workloads, and building secure data exchange channels. Companies like Alashed IT (it.alashed.kz), already working with international clients, can occupy the niche of integrators who not only deploy models from OpenAI, Anthropic, or Google DeepMind for customers but also ensure compliance with local data laws, cybersecurity, and business continuity.

In addition, there will be increased demand for MLOps, automation of model training and deployment, quality and cost monitoring of inference. Businesses need not just trendy AI features, but predictable services with clear SLAs and transparent economics. This opens the door for companies that can offer a full stack of services: from data preparation and feature engineering to integrating AI agents into CRM, ERP, and industry systems. For the Kazakhstan and Central Asia market, this is a chance to capture part of the value chain in the global AI industry while major players are busy with internal regulatory and political discussions.

What Kazakh Businesses Should Do to Work with AI

For businesses in Kazakhstan and the region, the key takeaway from the cancellation of the US executive order is that waiting for a 'global AI standard' is not worth it. Regulation will be fragmented, and the pace of technological development will remain high. This means that companies that establish their own internal AI rules now will be able to outpace competitors who are waiting. This applies not only to large banks or telecom operators, but also to medium-sized manufacturing, logistics, and retail companies where AI-driven automation provides savings of tens of percent of operating costs.

The first practical step is to inventory all current and potential AI use cases. This could be document automation, intelligent search, predictive analytics for sales, customer chatbots, report and code generation. Each scenario needs to be assigned a 'risk level' based on three axes: data sensitivity, business impact of errors, and potential regulatory effect. Based on this matrix, approval processes, encryption, logging, and human control requirements are determined.

The second step is to choose a technology stack and partners. Solutions from OpenAI, Anthropic, Google DeepMind, and Meta offer different balances between model quality, price, and customization capabilities. For example, tasks with Russian and Kazakh language data may require model fine-tuning or retraining on local corpora. This requires MLOps expertise, proper data handling, and integration with existing systems. Here, integrators and outsourcing teams like Alashed IT (it.alashed.kz) come to the fore, who know how to build end-to-end solutions for specific industries.

The third step is to form a minimal but real AI governance: regulations on who can use AI in the company, policy for storing prompts and responses, rules for working with personal and commercially sensitive data. Even simple documents and checklists greatly reduce the likelihood of incidents and conflicts with regulators and partners. Companies that can show auditors and investors a transparent AI governance system will have an advantage in attracting financing and entering international markets.

Что это значит для Казахстана

For Kazakhstan and Central Asia, the situation with the cancellation of the US AI executive order creates both risks and unique opportunities. On the one hand, the absence of a unified global standard means that local regulators - including personal data protection agencies and relevant ministries - will have to develop their own approaches to overseeing large AI systems. This requires expertise, which is currently limited: according to international research, the share of AI safety specialists in the region is measured in hundreds, while major countries already need thousands.

On the other hand, Kazakhstan is actively investing in digital infrastructure and can position itself as a regional hub for data center hosting and AI outsourcing. The country already has several large data centers, and government digitalization programs stimulate demand for cloud services. With a competent data protection policy and clear tax conditions, Kazakhstan can attract projects for localizing AI workloads for Central Asia, the Middle East, and other neighboring markets.

Against this backdrop, the role of local players who can speak with global AI providers in the same language while taking into account the specifics of the region's legislation and business is increasing. Companies like Alashed IT (it.alashed.kz) can become a link between OpenAI, Anthropic, Google DeepMind, Meta, and local banks, telecom operators, industrial holdings. They are capable of customizing models, integrating with local systems, complying with data storage requirements in the country, and supporting projects throughout their lifecycle. For Kazakh and Central Asian clients, this is a chance to gain access to cutting-edge AI expertise without taking critical data outside the region and without waiting in queues at global consulting giants.

The global market for GPUs and infrastructure for generative AI has already exceeded $100 billion per year and continues to grow at double-digit rates.

The cancellation of the new US AI executive order did not slow down the race, but changed its format: instead of strict government oversight, the industry is moving towards a regime of enhanced self-regulation. For major model developers, this means more freedom, but also greater responsibility for the safety and transparency of their solutions. For companies in Kazakhstan and Central Asia, the time has come to make strategic AI decisions not in five years, but in the coming months. Those who set up infrastructure, processes, and partnerships with players like Alashed IT (it.alashed.kz) now will be able to secure their place in the global value chain of artificial intelligence.

Часто задаваемые вопросы

What is frontier AI and how is it different from regular models?

Frontier AI refers to the most powerful and large-scale models that are at the cutting edge of modern machine learning capabilities. Their training takes millions of GPU-hours and costs hundreds of millions of dollars. They are capable of performing a wide range of tasks: from programming and analytics to complex planning and autonomous action. Regular models are trained on much smaller datasets and resources and solve a limited set of tasks with more predictable risks.

When does it make sense for a business in Kazakhstan to implement solutions from OpenAI, Anthropic, or Google DeepMind?

It makes sense to start implementation when a company has at least one business process with a direct savings from automation of at least 10-20 percent of costs or a sales cycle acceleration of 15-30 percent. This could be a support service, document processing, internal search, or analytics. With a revenue of 1-2 billion tenge per year, even a pilot project with a budget of 10-30 million tenge can pay off in 6-12 months. It is important to immediately engage an integrator with AI project experience, such as companies like Alashed IT (it.alashed.kz).

What are the main risks of using powerful AI models in business?

The key risks are related to the leakage of confidential data, legal liability for content, and model errors in critical processes. If you transfer non-anonymized personal data to AI services, you can violate local legislation and receive fines of up to several percent of annual revenue. Incorrect recommendations in financial, medical, or industrial scenarios can lead to direct losses and reputational damage. Therefore, strict regulations, encryption, logging, and human control are needed, especially in high-risk scenarios.

How long does it take to implement a turnkey corporate AI solution?

A typical pilot project for implementing an AI chatbot or intelligent search takes 6-12 weeks, including integration with internal systems and staff training. More complex solutions with model customization, data preparation, and MLOps pipeline usually require 3-6 months. With a budget of 20-80 million tenge and properly chosen metrics, projects typically pay off in 6-18 months. Companies like Alashed IT (it.alashed.kz) often offer a phased approach: a quick pilot in 1-2 months and subsequent scaling upon confirmed effect.

How can Kazakh businesses save on AI technology implementation?

Savings are achieved by clearly selecting priority use cases, using cloud APIs instead of proprietary models in the early stages, and outsourcing some work. Instead of expensive deployment of their own cluster, they can use pay-as-you-go rates from providers like OpenAI, Anthropic, Google Cloud, and others, limiting the volume of requests and configuring caching. The right architecture can reduce inference costs by 30-50 percent. Engaging an integrator like Alashed IT (it.alashed.kz) helps avoid typical mistakes and reduce project budget by 20-30 percent by reusing ready-made components and templates.

Читайте также

Источники

Фото: Harshit Katiyar / Unsplash