As artificial intelligence becomes more common in business settings, experts are warning that these tools could be the next major target for online criminals.
Some of the biggest software companies, like Microsoft and SAP, have recently started using AI systems that can handle office tasks such as finance and data management. But these digital programs also come with new security risks.
What Are These Digital Identities?
In today’s automated world, many apps and devices run tasks on their own. To do this, they use something called digital identities — known in tech terms as non-human identities, or NHIs. These are like virtual badges that allow machines to connect and work together without human help.
The problem is that every one of these digital identities could become a door for hackers to enter a company’s system.
Why Are They Being Ignored?
Modern businesses now rely on large numbers of these machine profiles. Because there are so many, they often go unnoticed during security checks. This makes them easy targets for cybercriminals.
A recent report found that nearly one out of every five companies had already dealt with a security problem involving one of these digital identities.
Unsafe Habits Increase the Risk
Many companies fail to change or update the credentials of these identities in a timely manner. This is a basic safety step that should be done often. However, studies show that more than 70% of these identities are left unchanged for long periods, which leaves them vulnerable to attacks.
Another issue is that nearly all organizations allow outside vendors to access their digital identities. When third parties are involved, there is a bigger chance that something could go wrong, especially if those vendors don’t have strong security systems of their own.
Experts say that keeping old login details in use while also giving access to outsiders creates serious weak spots in a company's defense.
What Needs to Be Done
As businesses begin using AI agents more widely, the number of digital identities is growing quickly. If they are not protected, hackers could use them to gain control over company data and systems.
Experts suggest that companies should treat these machine profiles just like human accounts. That means regularly updating passwords, limiting who has access, and monitoring their use closely.
With the rise of AI in workplaces, keeping these tools safe is now more important than ever.
Meta has introduced a fresh set of artificial intelligence models under the name Llama 4. This release includes three new versions: Scout, Maverick, and Behemoth. Each one has been designed to better understand and respond to a mix of text, images, and videos.
The reason behind this launch seems to be rising competition, especially from Chinese companies like DeepSeek. Their recent models have been doing so well that Meta rushed to improve its own tools to keep up.
Where You Can Access Llama 4
The Scout and Maverick models are now available online through Meta’s official site and other developer platforms like Hugging Face. However, Behemoth is still in the testing phase and hasn’t been released yet.
Meta has already added Llama 4 to its own digital assistant, which is built into apps like WhatsApp, Instagram, and Messenger in several countries. However, some special features are only available in the U.S. and only in English for now.
Who Can and Can’t Use It
Meta has placed some limits on who can access Llama 4. People and companies based in the European Union are not allowed to use or share these models, likely due to strict data rules in that region. Also, very large companies, those with over 700 million monthly users — must first get permission from Meta.
Smarter Design, Better Performance
Llama 4 is Meta’s first release using a new design method called "Mixture of Experts." This means the model can divide big tasks into smaller parts and assign each part to a different “expert” inside the system. This makes it faster and more efficient.
For example, the Maverick model has 400 billion total "parameters" (which basically measure how smart it is), but it only uses a small part of them at a time. Scout, the lighter model, is great for reading long documents or big sections of code and can run on a single high-powered computer chip. Maverick needs a more advanced system to function properly.
Behemoth: The Most Advanced One Yet
Behemoth, which is still being developed, will be the most powerful version. It will have a huge amount of learning data and is expected to perform better than many leading models in science and math-based tasks. But it will also need very strong computing systems to work.
One big change in this new version is how it handles sensitive topics. Previous models often avoided difficult questions. Now, Llama 4 is trained to give clearer, fairer answers on political or controversial issues. Meta says the goal is to make the AI more helpful to users, no matter what their views are.
The latest "Qwen2.5-Omni-7B" is a multimodal model- it can process inputs like audio/video, text, and images- while also creating real-time text and natural speech responses, Alibaba’s cloud website reports. It also said that the model can be used on edge devices such as smartphones, providing higher efficiency without giving up on performance.
According to Alibaba, the “unique combination makes it the perfect foundation for developing agile, cost-effective AI agents that deliver tangible value, especially intelligent voice applications.” For instance, the AI can be used to assist visually impaired individuals to navigate their environment via real-time audio description.
The latest model is open-sourced on forums GitHub and Hugging Face, after a rising trend in China post DeepSeek breakthrough R1 model open-source. Open-source means a software in which the source code is created freely on web for potential modification and redistribution.
In recent years, Alibaba claims it has open-sourced more that 200 generative AI models. In the noise of China’s AI dominance intensified by DeepSeek due to its shoe string budget and capabilities, Alibaba and genAI competitors are also releasing new, cost-cutting models and services an exceptional case.
Last week, Chinese tech mammoth Baidu launched a new multimodal foundational model and its first reasoning-based model. Likewise, Alibaba introduced its updated Qwen 2.5 AI model in January and also launched a new variant of its AI assistant tool Quark this month.
Alibaba has also made strong commitments to its AI plan, recently, it announced a plan to put $53 billion in its cloud computing and AI infrastructure over the next three years, even surpassing its spending in the space over the past decade.
CNBC talked with Kai Wang, Asia Senior equity analyst at Morningstar, Mr Kai told CNBC that “large Chinese tech players such as Alibaba, which build data centers to meet the computing needs of AI in addition to building their own LLMs, are well positioned to benefit from China's post-DeepSeek AI boom.” According to CNBC, “Alibaba secured a major win for its AI business last month when it confirmed that the company was partnering with Apple to roll out AI integration for iPhones sold in China.”