83% of artificial intelligence (AI) security breaches in manufacturing occur when employees share information with AI tools in violation of their company’s policies.
This is revealed by a new study by Netskope Threat Labs, which identifies this practice as one of the main security risks associated with the adoption of these technologies. The use of AI applications without proper safeguards can lead to the exposure of confidential documents and sensitive data of organizations.
According to the report, regulated data – that is, personal or financial information protected by legislation – is the data that appears most frequently in violations of corporate data protection policies related to AI. They represent 57% of recorded incidents. They are followed by passwords and API keys, with 19%; intellectual property, with 16%, and source code, with 9%.
These results highlight the risks involved in using artificial intelligence tools without adequate supervision, especially in a sector such as manufacturing, where technical information, operational data and intellectual property constitute strategic assets.
Manufacturers are making progress in controlling the use of AI
Despite the risks that information protection still poses, manufacturing companies are making progress in corporate control of artificial intelligence applications. Over the past year, the percentage of users using AI applications managed by their organizations has doubled, from 35% to 70%. In parallel, the use of personal accounts has decreased from 75% to 55%.
However, there is another trend that worries security managers: the percentage of employees who switch between personal and corporate accounts has increased from 9% to 25%. This behavior shows that the so-called “shadow AI” continues to spread, despite advances in business control. The use of unauthorized tools continues to generate risks related to loss of visibility, exposure of information and non-compliance with internal policies.
This evolution coincides with a rapid expansion of artificial intelligence in the manufacturing sector. The active use of these technologies has increased from 40% to 67% in the last year. In addition, its presence extends to common business applications: 96% of employees use software that incorporates AI capabilities and 90% interact with systems that use customer or user data to train models.
Adding to these risks is the growing connection of AI tools with other external systems and services. Automated connections to remote servers have increased by 490%, while the volume of interactions has increased by 600%. In this scenario, organizations need to know what information these tools can consult, how they process it, and where it is stored or shared. Monitoring these connections is essential to anticipate possible risks and prevent unauthorized access to corporate data.
Automated MCP connections increase
“Manufacturing organizations have gone far beyond initial testing with basic AI tools. “We are seeing a clear transition towards deep operational integration, driven by enterprise-managed solutions such as Microsoft Copilot and developer platforms such as Anthropic, along with a strong increase in automated MCP connections,” said Gianpietro Cutolo, Cloud Threat Researcher at Netskope.
For organizations, the challenge is to take advantage of its benefits without losing control over information, applications and connections.
The expert adds that, although this adoption generates significant efficiency improvements, it also raises new security challenges related to non-human entities and data governance. Risks that, in his opinion, security teams must proactively address.
Among the main conclusions of the study carried out by Netskope, the following data also stand out:
- Microsoft Copilot leads adoption: It is the most widespread AI application in the manufacturing sector, with a presence in 90% of organizations. They are followed by Anthropic Code Platform, with 89%, and ChatGPT, with 76%.
- Increases blocking of certain applications: Particular Audience is the AI tool most frequently restricted by manufacturing companies, at 45%. Next are Emergent (39%) and DeepSeek (36%).
- Malicious AI-related honeypots are growing: Activity went from about 25 users per 100,000 employees in February 2026 to around 100. It remained stable until May and skyrocketed again in June, when it reached almost 175 users per 100,000.
- Interactions with malicious links intensify: This activity began to increase in July 2025 and reached 150 users per 100,000 in October. After a second rebound that began in February 2026, it reached 200 per 100,000 users in March.
The data reflects that the expansion of artificial intelligence in the manufacturing industry is not only transforming work processes, but also expanding the surface of exposure to new threats. For organizations, the challenge is to take advantage of its benefits without losing control over the information, applications and connections that intervene in their operations.
