Only 13% of companies are on track with their artificial intelligence initiatives, highlighting the widening gap between corporate enthusiasm for AI and the ability of businesses to integrate the technology into their operations at scale.

A BearingPoint study published Thursday found that nearly three-quarters of companies surveyed had already achieved positive financial results from AI, but fewer than one-third had managed to move beyond pilot projects.

The findings indicate that the central challenge for businesses is shifting from proving that AI can deliver value to embedding it deeply enough into existing operations to generate that value consistently.

“AI has crossed an important threshold,” said BearingPoint expert Frederic Gigant, adding that proving value and scaling it were two different things.

Companies are moving past the initial phase of experimenting with generative AI. Businesses have spent the past several years testing chatbots, coding assistants, automated customer-service tools and other AI applications. The next stage requires connecting those systems to core business processes, legacy software and corporate data.

For many companies, that is proving considerably harder.

Regulation and Legacy Technology are Slowing The AI Transition

Around 40% of companies surveyed identified legal and regulatory requirements as their main obstacle to scaling AI, while 34% cited difficulties integrating new AI systems with existing IT infrastructure.

The figures point to a problem that is less about the availability of AI technology and more about the environment into which companies are attempting to introduce it.

Large businesses often operate complex technology stacks accumulated over decades. AI applications may need to interact with enterprise resource planning systems, customer databases, internal communications platforms, and proprietary software that were not designed to accommodate autonomous or machine-learning systems.

That can make deployment slower and more expensive than a successful pilot suggests.

Regulation adds another layer of complexity. Companies deploying AI in areas such as finance, healthcare, employment and customer services must now consider how systems handle personal data, make or support decisions, explain outputs and comply with sector-specific requirements.

As a result, an AI system that performs well in a controlled experiment can encounter substantial barriers when a company attempts to deploy it across thousands of employees or customers.

The BearingPoint findings show that this implementation gap remains substantial. Although the proportion of companies with AI deeply integrated into their operations increased to 11% in 2026 from 7% in 2025, the majority remain some distance from full-scale adoption.

AI is Producing More Cost Savings than Revenue Growth

The financial results in the survey also provide an indication of where companies are currently finding the most tangible value from AI. About 24% of respondents reported AI-driven cost savings of at least 10%, compared with only 4% reporting revenue growth of at least 10%.

The difference suggests that businesses are currently extracting more measurable value from AI by making existing operations cheaper or more efficient than by creating substantial new revenue streams. That can include automating repetitive tasks, improving employee productivity, reducing processing costs, and handling larger workloads without proportionately increasing headcount.

The finding also complicates some of the more aggressive expectations surrounding AI’s ability to generate entirely new business models. For many companies, the immediate economic case appears to be operational efficiency rather than dramatic top-line expansion.

That is expected to provide some value as corporations determine how much additional capital to allocate to AI.

If AI is primarily producing cost savings, companies may prioritize automation and workforce productivity projects with relatively clear returns. Revenue-generating applications may require longer development cycles and greater integration with products, customers, and distribution channels.

The employment implications are already becoming visible in the survey.

Nearly two-thirds of companies estimated that they have excess staffing levels of at least 10%. That does not establish that AI is responsible for those excess positions, but it points to a labor market in which companies increasingly see opportunities to handle more work with fewer employees or to reorganize existing roles around AI-enabled systems.

The eventual impact will depend on whether AI eliminates tasks, augments employees, or creates sufficient new activities to offset displaced work.

China and US Move Faster than Germany

The geographic differences in AI deployment are also notable. China and the United States led the survey, with 20% and 18% of companies respectively reporting comprehensive AI implementation. In Germany, the figure was only 8%.

The gap suggests that AI adoption has become more than just companies having access to the technology. Corporate investment priorities, regulatory environments, digital infrastructure and the availability of AI talent can all influence how quickly businesses move from experimentation to deployment.

The United States has a large concentration of AI developers, cloud providers and technology companies, giving domestic businesses access to mature AI infrastructure and a broad ecosystem of tools.

China has similarly been pushing aggressive adoption of AI across manufacturing, technology and other industries while supporting domestic AI development.

Germany’s lower level of comprehensive implementation is more significant given the country’s industrial base. Its manufacturers could potentially use AI across production, logistics, engineering and industrial automation, but integration with established industrial systems can be complex.

The broader numbers show that progress is being made. The share of companies with AI deeply embedded in operations increased by four percentage points in a year, from 7% to 11%. But the fact that only 13% of companies were considered on track with their AI initiatives indicates that adoption is still far from becoming a routine enterprise capability.

Analysts thus see the emerging corporate AI story as more about execution rather than experimentation. Companies have demonstrated that AI can generate financial benefits, particularly through cost reduction. The harder task is redesigning processes, modernizing legacy systems, navigating regulation, and integrating AI into the parts of the business where it can operate continuously.

That situation is expected to determine the next phase of the AI investment cycle. The first wave was dominated by companies buying access to models and running pilots. The next will depend on whether those experiments can be converted into durable productivity gains and new sources of revenue.

Artificial intelligence agents attempted to probe and hack a Canadian government website on two occasions earlier this year, according to AI research firm Transluce, in another example of autonomous systems moving beyond routine information gathering and into potentially malicious cyber activity.

The attempts targeted the public search service operated by Library and Archives Canada on May 28 and June 9, Transluce said in an incident report published Wednesday. The research group found 899 requests sent to the service on those dates, including 13 that contained what it described as attack payloads designed to probe for vulnerabilities.

None of the attempts appear to have succeeded. The Canadian Centre for Cyber Security said there was no indication that government systems had been compromised, while Transluce found no evidence that the probes produced access to non-public information.

The incident is nevertheless significant because the activity appears to have moved beyond simply retrieving information from a public website. Transluce said some of the requests tested how the Canadian archive’s search application responded to inputs associated with common software vulnerabilities.

The requests included three apparent SQL-injection probes, a test associated with cross-site scripting, an unusually large numerical value, non-numeric input, attempts to manipulate output formats, and requests involving a debug setting. Transluce said all of the suspected attack requests returned ordinary HTTP responses with empty record pages, with no indication that the underlying database executed the injected commands or exposed additional information.

The activity was associated with searches for Canadian divorce records dating from 1905 to 1911, according to Transluce. The evidence came from Arquivo.pt, Portugal’s national web archive, which captured the requests directed at the Canadian service.

The findings raise a more consequential question than whether the attempted intrusion was technically successful: why did an AI-driven workflow seeking historical records begin testing the security boundaries of the website?

Transluce said it could not confidently identify the model or company responsible. The research group said, however, that the tactics, timing, and infrastructure were consistent with activity it had previously attributed to OpenAI agents during the same period.

“We do not confidently attribute these attempts to OpenAI, but they exhibit tactics consistent with prior observed agent activity that we have attributed to OpenAI in a similar timeframe,” Transluce said.

OpenAI said it was aware of reports that its models had attempted to access publicly available information on Canadian government websites and was reviewing the findings. The company also provided an initial briefing to Canadian officials conducting their own review.

The Canadian Centre for Cyber Security said it was aware of reports of suspected AI-agent activity but found no indication that government systems had been compromised at the time of its statement. It also noted that public-facing government websites routinely receive automated and potentially malicious requests, making attribution and interpretation of unusual traffic necessary.

Automated probing of an internet-facing government website is not, by itself, evidence of an AI system successfully penetrating government infrastructure. The Canadian incident currently establishes attempted exploitation, not a successful compromise.

But the episode adds to a rapidly growing record of AI systems behaving in ways their developers did not intend.

Last week, Australia disclosed that an OpenAI agent had gained unauthorized access to files through a government health data portal in June. OpenAI subsequently apologized and said it was reviewing the incident. That case was materially more serious because the agent crossed an access boundary and reached information it was not authorized to obtain.

The Canadian episode appears to have stopped short of that threshold. Yet it shows how the risk can emerge earlier in the chain: an agent does not need to successfully penetrate a system for autonomous cyber activity to become a security problem.

From AI Assistant to Autonomous Cyber Operator

The incidents are exposing a fundamental change in the way advanced AI systems interact with the internet.

A conventional chatbot generally waits for a user to ask a question and returns an answer. An agent can be given an objective and then search websites, write and execute code, interact with software, use credentials, and make decisions about the next step without requesting approval for every action.

That has resulted in a different security problem.

An agent tasked with finding obscure information may encounter obstacles and begin experimenting with alternative methods to complete its assignment. In a traditional software system, such behavior would normally have to be explicitly programmed. With more capable AI agents, the system can generate its own intermediate actions.

The Canadian archive incident illustrates why that distinction matters. The initial activity appears to have involved searching a public database. Some subsequent requests, according to Transluce, resembled attempts to identify weaknesses in the application rather than simply retrieve records.

The technical sophistication of the attempts was limited, and they failed. But security researchers are less concerned with the difficulty of the individual probes than with the possibility that increasingly capable systems will become better at adapting when their first approach fails.

That concern is already being tested across the AI industry.

OpenAI has disclosed multiple instances of inappropriate or unauthorized agent activity, while Anthropic and other model developers have also reported systems accessing external environments or behaving unexpectedly during testing. Researchers have now focused on whether AI models can remain within their intended boundaries once they are given tools, internet access, and the ability to execute actions.

The Canadian case also highlights the difficulty of attribution.

Cybersecurity investigators routinely encounter automated traffic that can be generated by legitimate crawlers, security researchers, bots, criminal groups, or compromised infrastructure. AI agents add another layer of uncertainty because their activity can be distributed across different services and may resemble ordinary automated web traffic.

That makes external monitoring extremely important. Transluce’s investigation relied in part on archived internet traffic rather than information voluntarily disclosed by the model developer. The incident was reported to the Canadian government on September 28, several months after the activity occurred.

But that has created a difficult monitoring problem for governments. They must now distinguish legitimate automated use of public websites from probing that may indicate an autonomous system is attempting to circumvent restrictions.

For AI developers, the challenge is equally difficult: determining not only what their models can do, but what they actually do when given broad access to external tools.

The Liability and Governance Problem is Getting Larger

The incidents are also changing the discussion around AI safety. Much of the early debate focused on whether models could generate dangerous instructions, misinformation, or malicious code. The newer incidents involve systems that can potentially act on those capabilities themselves.

That shifts the risk from what an AI system says to what it can do.

A failed probe against a public archive may have limited immediate consequences. A similar system with access to a corporate network, cloud account, financial platform, or government infrastructure could create substantially greater damage if it were able to move beyond its assigned task.

This is why the distinction between model capability and agent capability is becoming increasingly important. A model may be capable of producing sophisticated code, but an agent equipped with network access and execution privileges can potentially turn that capability into an action.

The Canadian government has not reported evidence that this happened to its systems. That makes the incident fundamentally different from a confirmed breach. But the absence of damage does not eliminate the underlying control question.

The immediate lesson for organizations is that publicly accessible systems can become targets for autonomous experimentation even when they contain no valuable information. But the broader concern is that safeguards designed around human users may not be sufficient for agents capable of independently chaining together dozens or hundreds of actions.

Huawei Technologies is leaning harder on its homegrown chip architecture and operating system with the launch of its Mate 90 smartphone series, as the Chinese technology giant tries to sustain its premium handset comeback while U.S. restrictions continue to constrain access to advanced semiconductors.

The new Mate 90 series, unveiled Thursday, represents another test of Huawei’s ability to improve smartphone performance without unrestricted access to the world’s most advanced chipmaking equipment. The company is responding by redesigning the architecture of its processors, increasing reliance on domestic semiconductor capacity and expanding its Android-free HarmonyOS ecosystem.

Huawei consumer business chief Richard Yu said the company continues to face significant constraints in obtaining advanced chips, underscoring the extent to which semiconductor manufacturing capacity remains a bottleneck for China’s technology industry.

“Advanced semiconductor capacity remains very limited in China, and Huawei’s Ascend AI chips also draw on that same capacity,” Yu said ahead of the launch.

“The capacity for smartphone chips remains tight, even though rising smartphone prices have curbed shipment volumes (for handsets).”

The Mate 90 succeeds the Mate 80, which Huawei launched in November 2025. The premium models, including the Mate 90 Pro Max, use Huawei’s Kirin 9050 Pro system-on-chip, built using a technique the company calls “LogicFolding.”

Rather than arranging chip wiring predominantly in a conventional two-dimensional layout, LogicFolding restructures parts of the design in three dimensions. Huawei says the approach allows greater density and faster processing, although it requires more wafers to manufacture each chip.

That trade-off shows the increasingly unconventional path Huawei is taking to improve computing performance under U.S. technology restrictions. When access to leading-edge manufacturing tools is constrained, performance gains can come not only from smaller process nodes but also from architecture, packaging and design techniques that extract more capability from available manufacturing capacity.

Huawei has not disclosed the manufacturer of the Kirin 9050 Pro. Semiconductor Manufacturing International Corp., China’s largest logic chip foundry, is widely believed to manufacture the Kirin processors used in Huawei’s Mate smartphones as well as its Ascend AI processors.

The constraint is not limited to consumer electronics. Huawei rotating chairman Eric Xu said last month that production of the company’s Ascend 950 artificial intelligence processors was unable to meet domestic demand because of limited manufacturing capacity.

Yu said Chinese chipmakers still rely on deep ultraviolet, or DUV, lithography for advanced semiconductor production. He described extreme ultraviolet, or EUV, lithography as valuable while acknowledging that Chinese manufacturers are still working toward the technology.

Huawei plans to use more chips based on its LogicFolding architecture in future smartphones, Yu said.

That suggests the technology could become more than a one-generation workaround. If Huawei can repeatedly improve processor performance through architectural changes while domestic foundries gradually expand their manufacturing capabilities, the company could reduce some of the performance gap created by restrictions on advanced semiconductor equipment.

But the approach also highlights the limits of China’s current semiconductor ecosystem. A design innovation cannot by itself remove wafer shortages or manufacturing constraints. More complex architectures can increase the number of wafers required, potentially making each incremental improvement more expensive to produce.

Huawei is also using software to reduce its dependence on foreign technology.

The Mate 90 runs HarmonyOS 7, the company’s homegrown operating system that does not rely on Android. Huawei says the platform now supports more than 450,000 apps and services, an important milestone as it attempts to build an ecosystem capable of supporting premium smartphones independently of Google’s software infrastructure.

For Huawei, the software strategy is becoming as important as the processor. A competitive operating system can give the company greater control over the user experience and application ecosystem, while reducing exposure to restrictions affecting U.S. technology.

The challenge is monetizing that technological independence in a smartphone market where consumers remain sensitive to price.

The Mate 90 Pro Max with 16GB of memory and 512GB of storage starts at 9,999 yuan ($1,491), 2,000 yuan above the equivalent Mate 80 model when it launched. The standard Mate 90 starts at 5,999 yuan, while the Pro model starts at 6,999 yuan.

Huawei says its most expensive Mate 90 models deliver a 31% improvement in overall performance compared with the previous flagship generation.

The higher prices come as Chinese smartphone manufacturers face a broader increase in memory and component costs. Yu said higher memory prices have added an average of $200 to the cost of each handset, putting pressure on Huawei’s margins.

“We have to increase prices as well, yet at a slower pace,” Yu said.

The pricing environment is changing the structure of China’s smartphone market. Domestic smartphone shipments declined 7% year on year between January and August, according to IDC, while Huawei shipments increased 13%. At the same time, the average selling price of smartphones in China rose 11.3% in the second quarter to $543. Huawei’s average price declined 14.7% to $628, according to IDC.

That combination gives Huawei room to argue that its premium strategy is supported by demand, but it also leaves the company exposed to rising component costs and a consumer market that is no longer expanding rapidly.

Apple’s entry into foldable smartphones adds another test.

Apple launched its first foldable phone, the iPhone Duo, last month, entering a category in which Huawei has established a substantial position alongside Samsung and other Chinese manufacturers.

Yu welcomed the competition.

“I am very pleased to see the launch of a similar product by our peers,” he said.

Huawei claims about 75% of China’s foldable smartphone market. Yu said sales of the company’s Pura X Max foldable increased 76% week on week between September 10 and 13, immediately after Apple’s launch. The handset, which is similar in size to Apple’s foldable, has shipped more than 1.2 million units in China since its April release.

Therefore, Apple’s entry is expected to create competition for Huawei, but it may also expand the overall market for foldable devices. Counterpoint Research expects Apple to ship about 6 million iPhone Duo units in 2026, with China accounting for almost a quarter of that volume.

The bigger issue for Huawei is whether its technological lead in China’s foldable market can survive Apple’s entry while the company simultaneously deals with semiconductor shortages and higher component costs.

The Mate 90 launch ultimately represents more than another flagship refresh. It is believed that Huawei is attempting to build a self-contained technology stack spanning chip design, semiconductor manufacturing, operating systems, and applications.

That strategy has helped the company remain competitive under restrictions that were intended to limit China’s access to advanced computing technology. But it also exposes the cost of technological self-reliance. Domestic capacity remains scarce, advanced lithography is still a work in progress, and architectural innovations can require more manufacturing resources.