#Belgium #AI Security #Galahad #Corbenic AI #Oostkamp

Galahad Launch: A Critical Innovation for AI Security

As artificial intelligence (AI) applications become more prevalent, ensuring their secure operation is more important than ever. Recently, Belgian startup Corbenic AI introduced Galahad, a beta service aimed at providing robust protection for AI memory while enabling developers to trace and troubleshoot issues effectively.

Addressing Growing Concerns

Recent incidents reported by major AI players like OpenAI and Anthropic highlighted the urgent need for stronger security measures. OpenAI raised fears when their research agents unintentionally uploaded user images to unregulated websites, while Anthropic's systems accessed real corporate networks during testing. Such operations remind us how crucial it is to implement safeguards that prevent unintended data exposure when AI systems operate independently.

What is Galahad?

Designed specifically to meet these challenges, Galahad allows developers to protect AI's temporary memory where sensitive information, such as customer contracts, could reside. Corbenic AI has stated that the system secures stored information using encryption techniques and segregates data for each client, thereby minimizing the risk of unauthorized access.

"Galahad offers a secure vault for working memory along with an accurate log of all activities performed. It centralizes security concerns and audits, making it easier to track information flow within AI systems,” notes Sietse Schelpe, founder and CEO of Corbenic AI.

Energy Efficiency and Usability

One of the standout features of Galahad is its energy efficiency. The tool intelligently saves completed AI work, allowing users to revisit documents without the need to rerun extensive processing tasks. Testing within the company has shown that this can lead to an impressive reduction in graphics processor energy consumption by 92 to 93 percent, a significant advantage for enterprises aiming to cut down on operational costs.

By enabling a logging feature, developers are afforded the ability to scrutinize past activities and restore previous states of a memory file, consequently enhancing their capacity to mitigate errors, although it won't revert actions already taken, like sent emails.

Compatibility and Licensing

Galahad is not a standalone solution but rather integrates seamlessly with other popular AI-serving software, including vLLM, SGLang, and llama.cpp. Currently, the beta version supports a free, renewable license for non-commercial use across systems equipped with a single graphics processor (GPU), providing an accessible entry point for developers. For commercial entities, paid licensing options are available.

Conclusion

Corbenic AI's roll out of Galahad marks a critical intervention intended to strengthen the security of AI memory amidst a landscape of increasing automation and risk. As developers look to innovate while safeguarding user data, tools like Galahad will undoubtedly become indispensable in the evolving ecosystem of AI applications. Given the implications of its capabilities, this can pave the way for safer and more effective AI interactions in various sectors.

For those interested, more information can be accessed through Corbenic AI’s official channels and their dedicated GitHub page for Galahad’s beta.

For business inquiries, please contact:
Sietse Schelpe
Phone: +32 493 881 164
Email: [email protected]