Secure AI Systems
As AI services proliferate, companies and individuals are faced with an increasing complexity of providers, microservices, and terms of use.
AI Services drawbacks
When we integrate AI tools into our workflows, they can become vulnerable to activity spying, which can be easily exploited by attackers. The potential malicious uses of private data passed to cloud AI systems are alarming, including:
- Theft of personal information
- Work behavior and responsibility monitoring
- Intellectual property risks
- Enhanced social engineering cyberattacks
Inadequate Security Measures for security or experience
While measures like anonymous sessions, encrypted data, no-data-save policies, and end-to-end encrypted communications are crucial, they are insufficient to protect us. There is more than cybersecurity threats to using cloud system and most importantly use them FREELY in the cloud. Businesses will try to monetize the service.
We can elaborate on undesirable uses that AI companies could adopt (ads, tracking, training, etc.). But beyond annoying not having a history is, it’s also very sad to delete such precious usefull data for an AI because of the trust of third-party.
Unwanted uses might be for instance
- No-data-save/No-history policies decrease the value of AI systems for many tasks that require iterations
- Feed the training of a model
- Anonymous sessions can be compromised if the data contains personal information
- Encrypted data can still be vulnerable if the system is compromised
The Power of Local AI Systems
“BE LOCAL (and open !)” is a mantra that is changing the way we approach AI development and deployment.
By keeping AI systems local, we can significantly reduce the risk of data leakage and confidentiality breaches. Sensitive information will not be transmitted to an obscur third-party, where it can be intercepted or accessed by unattended parties.
The Hardware Challenge
Currently, an AI system requires a significant investment, with professional equipment costing around $5,000. However, as AI systems develop and computation becomes a standard for modern computers, it will become easier to run local AI systems at a lower cost. Prices are likely to drop by a factor of two every two years for the same amount of performance.
While the benefits of local AI systems are clear, their development requires financial support from transparent and independent organizations. The open-source community for local AI development needs resources to build great, secure local AI systems.
Conclusion
The development of secure AI systems is a pressing concern that requires immediate attention. By promoting local AI systems, we can create a more resilient and secure ecosystem that benefits individuals, organizations, and communities alike.
Sources: