Netskope’s report finds manufacturers’ AI use is outpacing governance, with rising data leaks, agent connections and AI-themed attacks.
Manufacturers are adopting AI at speed, and attackers have noticed. That is the central message of the Netskope Threat Labs Report on manufacturing, published in November 2026. It draws on anonymized telemetry from the Netskope One platform, covering June 2025 to August 2026, rather than surveys.
Adoption is surging
Active AI users in manufacturing rose from 40% to 67% in a year. Use of organization-managed AI tools doubled from 35% to 70%, while use of personal AI accounts fell from 75% to 55%. That suggests companies are steering staff away from “shadow AI”. Even so, the shift is not complete.
The shift is not complete. Users switching between personal and enterprise accounts jumped from 9% to 25%, indicating that employees still experiment with tools outside the approved ones. Netskope advises faster approval of new AI apps, tighter instance-level controls, and a simple request process.
Copilot overtakes ChatGPT
For the first time, Microsoft Copilot is the most widely adopted AI application in the sector, used by 90% of organizations. Claude Platform follows at 89% and Claude Code at 84%, while ChatGPT trails at 76%. The report notes that this differs from the global picture, reflecting the fact that AI now sits within business software.
Much of the AI traffic now runs within existing business software and development environments, not in standalone chatbots. About 96% of employees use apps with embedded AI features, and 90% interact with systems that use customer or user data for training. That makes tracking sensitive data harder and raises security concerns.
Sensitive data is leaking
Regulated data makes up 57% of AI-related data policy violations. Passwords and keys account for 19%, intellectual property 16%, and source code 9%. Upstream violations, meaning sensitive data entered into AI tools, make up 83% of all AI-related violations. For manufacturers, intellectual property such as designs and process know-how is especially valuable, which raises the stakes.
Agents are connecting to the outside world
The report flags a sharp rise in agentic AI. Non-human entities connecting to remote MCP servers grew 490%, and MCP events rose 600%, with a spike in July and August. Each connection gives an AI agent a path to a server outside the company, so security teams must know which agents connect and what data they carry.
Attackers follow the trend
Malicious AI lures that pose as trusted AI brands to deliver malware rose from around 25 to roughly 100 users per 100,000 employees in February. They then reached nearly 175 per 100,000 in June. Campaigns include fake AI installers and Trojanized developer tools, showing how attackers follow the same trend.
Malicious links returned by AI apps also come in waves. They reached about 150 per 100,000 users in October 2025 and about 200 per 100,000 users in March 2026.
Companies are also blocking risky apps. Particular Audience is blocked by 45% of organizations, Emergent by 39%, and DeepSeek by 36%.
What Netskope recommends
- Inspect all web and cloud downloads.
- Allow reputable apps and block the rest.
- Use data loss prevention (DLP) policies for sensitive data.
- Use remote browser isolation for risky sites.
- Gain visibility and control over AI apps, agents and API traffic.
The bottom line
Manufacturing faces higher stakes than most sectors, because a cyber incident can halt a production line. The key takeaway is that AI is now embedded in everyday tools, so governance can no longer rely on blocking a few chatbots. Visibility into which AI apps and agents operate, and what they can reach, is now the baseline.
Note: the report comes from a security vendor and its recommendations point to Netskope’s own products, so readers should weigh it accordingly.
