Bastion Launches Security Assessments for Enterprise AI Models
Demand for these services in the Russian market is expected to grow rapidly over the next several years.
Russian companies are adopting artificial intelligence at an accelerating pace. Large language models can take over a substantial share of routine business tasks. At the same time, broader adoption has been slowed by concerns among many business leaders about the security of these systems. A new offering from Bastion aims to address that challenge.
Comprehensive Risk Assessment
The information security company has expanded its service portfolio. Bastion now offers security assessments for enterprise AI systems as well as hardware-software systems through its Hardware Hacking service.
Testing enterprise AI models for vulnerabilities enables organizations to assess the current security posture of their internal LLMs. The company's security professionals evaluate the protection of all processed data and the resources required for corporate neural networks to operate reliably. The service also measures the likelihood that confidential information could be exposed through AI-generated outputs, helping organizations manage the risk of data leakage through internal services.
Security assessments for hardware-software systems include comprehensive testing of smart devices, industrial systems, and electronic equipment to identify weaknesses in their security posture. Bastion's team evaluates both hardware and embedded software by simulating real-world attack scenarios, giving customers a clear understanding of the potential impact of a cyber incident.
A Broad Customer Base
According to company representatives, the new services are expected to appeal to a broad range of organizations, particularly medium-sized and large enterprises. Those organizations account for most deployed enterprise AI models and hardware-software systems. Healthcare providers, financial institutions, and retail companies have especially strong incentives to secure their hardware-software systems because the risk of compromise is particularly high in those sectors and the consequences of failure can be severe.
"Many large organizations now rely on proprietary LLMs in their day-to-day operations. These models contain significant volumes of internal corporate information, and compromising that data can create serious risks ranging from information leaks to business process disruption. That makes protecting LLMs an important and increasingly sought-after area of cybersecurity," said Dmitry Kalinin, Head of the Vulnerability and Cyber Incident Department at Bastion.
Responding to Market Demand
Bastion's new services clearly reflect the current direction of the market. Cybersecurity professionals broadly agree that security assessments for enterprise neural networks and hardware-software systems will soon become a standard practice rather than a specialized offering. Organizations are likely to test these environments in much the same way they already assess web applications, mobile services, and other components of their IT infrastructure.
The security of enterprise neural networks and hardware-software systems has been under discussion for several years. As early as 2024, cybersecurity researchers identified a number of key risks associated with AI systems, including exposure of personal data and trade secrets, leakage of information from training datasets, manipulation of model outputs, and the use of AI models to generate vulnerable code. Meanwhile, the growing adoption of hardware-software systems has highlighted the need to evaluate not only software security but also hardware compatibility, configuration, and the security of the underlying hardware platform.
AI Security Continues to Advance
In 2025, Russia introduced its first domestically developed cybersecurity system designed to protect AI models from cyberattacks. The solution was created by graduate students at ITMO University together with experts from Raft. It is compatible with all major AI models, regardless of whether they are open or proprietary, allowing organizations to tailor security policies to their own operational requirements.
The market for specialized security technologies designed to protect large language models continues to expand rapidly. These products are expected to become more sophisticated while the range of available solutions continues to grow. Over time, organizations may shift away from one-time security assessments toward continuous, comprehensive security audits.
