Ensuring the security of the AI ​​assistant

[rakhirhif8963]

“It’s the combinatorics here that should make everyone worry,” he says. “These categorical risks exist in the broader technology based on natural language models, and when you combine them with the runtime security risks that we’ve already seen, as well as the information access and auditability risks, it ends up having a multiplier effect on risk.”

Desktop AI is likely to go mainstream in 2025. Companies are already racing to quickly adopt Microsoft 365 Copilot and other desktop AI technologies, but only 16% have moved from early pilots to rolling out the technology to all employees, according to a Gartner study. The vast majority (60%) are still evaluating the technology as a pilot, and a fifth of enterprises have not yet reached that point and are in the planning stages.

Most workers are looking forward to having a desktop AI system to help them with their daily tasks. According to Gartner, nearly 90% of respondents believe their users will struggle to access their AI assistant, and 89% agree that the technology improves productivity.

Unfortunately, these technologies are “black boxes” in terms of their architecture and security, meaning they lack trust. Alcove says companies can run background checks on personal assistants, restrict their bahamas mobile database to certain technologies, and audit their performance — measures that currently have no equivalent in desktop AI systems.

AI assistants — whether on a desktop, mobile, or in the cloud — will have far more access to information than they need, he says. “Current technology is not ready to handle the fact that my assistant can do a certain set of electronic tasks on my behalf, and that’s it,” Alcove laments. “You can give your assistant access to your email and calendar, but you can’t prevent them from seeing certain emails and calendar events. They can see everything.”

According to him, such an opportunity to delegate tasks should become part of the security system of AI assistants.

Cyber ​​risk: social engineering targeting both users and AI
Without such a security and control system, attacks are likely to follow.