BlackFog ADX Vision 2.0 Bolsters Enterprise AI Security with Prompt Protection and Governance
BlackFog's ADX Vision 2.0 enhances enterprise AI security by introducing robust governance and control mechanisms for generative and agentic AI, addressing risks like prompt injection and unauthorized data exposure.
BlackFog has launched ADX Vision 2.0, an updated iteration of its AI security platform designed to provide organizations with enhanced governance and control over the deployment and use of generative and agentic artificial intelligence.
As AI evolves from simple user interaction tools to autonomous agents capable of making decisions and acting upon corporate data, traditional security frameworks are proving insufficient. New threats such as prompt injection, unauthorized data exfiltration, and a lack of visibility into AI's data handling practices have emerged, necessitating specialized security solutions.
ADX Vision 2.0 operates as an independent control layer directly on the endpoint, intercepting data and AI interactions before they reach external AI services. This approach grants organizations comprehensive visibility, governance, and auditability for all AI activities, whether initiated by human users or AI agents. It enforces policies dictating how data can be utilized by any AI tool or model.
The new version builds upon ADX Vision's existing capabilities in Shadow AI discovery and control, expanding its scope to address the growing complexities of autonomous AI systems within the enterprise.
A key feature of ADX Vision 2.0 is its introduction of seven distinct layers of prompt protection. These layers are engineered to detect and thwart prompt injection attempts and other malicious or manipulated inputs. They include protections against context switching, obfuscation, jailbreaking, data exfiltration, dangerous prompt patterns, fuzzy keywords, and structural anomalies. These defenses are applied uniformly whether the prompt originates from an employee or an AI agent acting on their behalf, extending security to automated workflows lacking direct human oversight.
Furthermore, ADX Vision 2.0 allows for the implementation of custom controls at company or departmental levels. These controls can identify and safeguard sensitive information, such as credit card numbers, Social Security numbers, and proprietary domain-specific terminology, ensuring it is not inadvertently exposed through AI interactions.
To bolster control and auditability, the platform enables organizations to redirect AI requests to sanctioned, officially licensed LLM services. This ensures corporate data remains within approved environments while still allowing employees to leverage AI. An optional prompt auditing feature for leading frontier LLMs provides deeper insights into information shared with AI systems, supporting compliance and governance requirements.
"AI is quickly moving from something employees interact with to something that can make decisions and take action on their behalf," stated Dr. Darren Williams, CEO of BlackFog. "That changes the security equation. When an agent is sending prompts on an employee’s behalf, there’s no human in the loop to catch a manipulated instruction or a leak of sensitive data. ADX Vision 2.0 inspects agentic prompts with the same protections as those written by humans."
Finally, ADX Vision 2.0 introduces unified visibility into token consumption across various LLMs. This feature aids in financial planning and cost management, while also helping to identify potential token theft, where attackers exploit stolen credentials to incur unauthorized usage costs.