Lasso Security brings contextual data protection to generative AI applications

Lasso Security Co., Ltd.a generative artificial intelligence security company, today announced the launch of a custom context policy wizard aimed at helping enterprises avoid data breaches when using everyday tools such as OpenAI's chatbot ChatGPT.

The company provides big Language model cybersecurity and data management for enterprises through an end-to-end solution. Discover what AI apps and tools employees are using, and help managers create policies around their use to prevent data and knowledge leaks.

With the proliferation of AI tools in the workplace, data management is becoming increasingly complex.Although more third-party large-scale language model services are introducing security and privacy rules regarding the data passing through them, important To ensure thorough internal compliance be Ensure that employees are not accidentally using prompts that should not be used in external tools.

Previously, this protection was provided by rule-based policies that used patterns to detect problematic prompts given to LLM.,but, Employees can inadvertently circumvent the pattern by filling in the prompts. Lasso's new custom policy wizard integrates with the company's browser extensions and secure gateways, allowing administrators to set policy guidelines in plain English.

“It’s all about context,” said Ophir Dror, co-founder and chief product officer at Lasso. safety, told SiliconANGLE. “To solve the new problem of knowledge leakage (rather than structured data leakage)” we The way we look at data protection has completely changed. No more patterns or predefined regular expressions that can't catch anything “near” or “similar.” ”

For example, if a policy prohibits HR professionals from discussing pay, the AI ​​engine will understand and block the interaction. the story About wages and compensation and Benefits of Organizational situation. However, it is possible to talk about general and public knowledge about salaries in your company.

Policies set by an administrator remain in effect and can be changed by the administrator at any time. Administrators receive alerts and telemetry about policy performance. parallel Verification process is performed creation of No matter what the rules are. There are also ways to test your policies against different use cases and the adjustment processes that keep your employees productive. They are Under adjustment.

From the user's perspective, There are several options that an administrator can set Based on your organization's policies. In most cases, if a problem occurs, foundthe session is simply Blocked, user must create new prompt to do move on. Administrator Alerts generated and An entry appears in the admin console To give permission An administrator will investigate.

“In the age of generative AI, traditional data protection mechanisms are no longer sufficient,” Dror says. “Structured data remains a concern, but a new concern is emerging: knowledge leakage. am sending Specification of the next feature, the designer will send an overview of the future model and When finance people submit their budgets, the existing security stack breaks.”

Image: Lasso Scality