Generative AI is advancing rapidly, but so are imaginative techniques men and women uncover to use it maliciously. Several governments are attempting to pace up their regulating plans to mitigate the risk of AI misuse.
In the meantime, some generative AI builders are on the lookout into how they could assistance protected their models and products and services. Google, owner of the generative AI chatbot Bard and father or mother company of AI investigation lab DeepMind, launched its Secure AI Framework (SAIF) on June 8, 2023.
SAIF is established to be “a bold and accountable, […] conceptual framework to help collaboratively protected AI technology,” Royal Hansen, Google’s VP of engineering for privateness, security and security, and Phil Venables, CISO of Google Cloud, wrote in a launching paper.
The effort and hard work builds on Google’s expertise developing cybersecurity styles, these as the collaborative Source-chain Ranges for Application Artifacts (SLSA) framework and BeyondCorp, its zero believe in architecture employed by a lot of organizations.
Precisely, SAIF is “a initial step” made to support mitigate dangers specific to AI programs like theft of the model, knowledge poisoning of the schooling facts, destructive inputs by prompt injection and extracting confidential facts in the coaching knowledge.
SAIF is designed all around 6 main principles:
“We will quickly publish quite a few open-resource equipment to aid place SAIF features into exercise for AI security,” Hansen and Venables stated.
They also vowed to increase Google’s bug hunter courses to reward and incentivize analysis about AI safety and security.
Browse more: Moral Hackers Could Gain up to $20,000 Uncovering ChatGPT Vulnerabilities
Eventually, they reported that Google was committed to helping produce intercontinental criteria on AI security, these as the US National Institute of Requirements and Technology’s (NIST) AI Risk Management Framework and Cybersecurity Framework, as properly as ISO/IEC 42001 AI Management Technique and ISO/IEC 27001 Security Management System requirements.
Some parts of this article are sourced from:
www.infosecurity-journal.com