Artificial intelligence (AI) is transforming society and business, promising extraordinary advances in productivity and innovation. However, the security of these systems has not kept pace. The recent case of DeepSeek, a Chinese startup that publicly exposed a database containing sensitive information, highlights a critical issue: we are building a new digital era on fragile foundations, prioritising innovation over security.
DeepSeek gained recognition for its advanced models but revealed a concerning vulnerability by exposing user conversation histories, secret API keys, and backend service metadata. This incident is just one example of what can go wrong when basic security measures are overlooked. The absence of fundamental controls not only exposes technical gaps but also reflects a governance failure in managing AI-related risks.
This is not an isolated case—many companies focus on the scalability and performance of their AI models without embedding essential cybersecurity principles. The result is vulnerable systems prone to attacks, manipulation, and data breaches, which can compromise both user privacy and the integrity of the models.
AI security cannot be treated as a mere by-product of innovation. There is a significant gap between identifying problems and implementing effective measures, making the role of both companies and users essential in mitigating risks. Providers must adopt a Zero Trust approach, from development to deployment, while users should conduct risk assessments before adopting new AI-based solutions. This includes setting rules for the retention and deletion of sensitive data, implementing privacy-preserving techniques, and establishing governance frameworks to monitor data collection and usage.
The DeepSeek case serves as an urgent warning: either we integrate security into AI now, or we will face a future filled with avoidable risks and crises. Only those who successfully balance innovation with cybersecurity will be prepared to lead this new technological revolution responsibly and sustainably.
Read the full case here.
