Through the Coalition for Secure AI (CoSAI), tech giants and AI developers want to make AI applications more secure by developing specific tools and building an ecosystem to share best practices.
Security of AI applications is becoming increasingly important, and the technology industry wants to take responsibility for this. Several technology giants and AI developers want to work together to address this security, and so they founded the CoSAI partnership. The CoSAI partnership operates under the banner of the non-profit OASIS Open, an organization that works with various open source projects, especially in the field of cybersecurity.
Participants include the big three hyperscalers AWS, Microsoft and Google, as well as AI developers such as OpenAI, Anthropic, Cohere and GenLab. Other participants include well-known technology companies such as Nvidia, Intel, IBM, Cisco, PayPal, Wiz and Chainguard.
The CoSAI collaboration will complement AI development initiatives already underway. Work within this new initiative should be complementary, and thus will take into account work from OpenSSF, the AI Alliance, the Cloud Security Alliance, Partnerships on AI, and the Frontier Model Forum.
Two goals
This new partnership has two goals: first, to provide businesses and organizations with the tools and technical expertise they need to secure AI applications, and second, to create an ecosystem where companies can share AI-related cybersecurity best practices and technologies.
To these ends, three open source “workstreams” or projects have been initiated. The first project is to help software developers scan their ML workloads for security risks. To this end, a taxonomy of known vulnerabilities and solutions that address them is being developed. Additionally, CoSAI participants also want to develop a “cybersecurity scorecard” to monitor vulnerabilities in AI systems and report them to stakeholders.
The second project of this partnership focuses on addressing AI security risks, with the goal of identifying investments and transition approaches that may have security implications for the use of AI.
The third initiative CoSAI has launched is to pay attention to the risks of supply chain attacks, for example using software components from public repositories and libraries such as GitHub. Checking these software components for vulnerabilities is often a long and complicated process. CoSAI now wants to develop workflows that simplify these processes.
Instilling fear in open source
At the same time, the Alliance will also pay close attention to the risks of using third-party AI models to develop solutions and applications. Consider open-source neural networks as an alternative to building costly proprietary algorithms. However, these open-source models may actually pose vulnerabilities that can be exploited by hackers.
CoSAI members plan to launch many more projects in the future. All efforts will be overseen by a technical committee made up of AI experts from academia and the private sector.
Read also: NYT: Hackers steal details of OpenAI AI technology
