BNP Paribas, Mistral and TELUS Digital Highlight Mounting Concern Over AI Cybersecurity Risks

Concern is growing over the ability of advanced artificial intelligence models to uncover cybersecurity vulnerabilities, prompting new defensive efforts and calls for stronger testing of enterprise AI systems.

BNP Paribas has partnered with AI startup Mistral to improve its defenses against potential threats linked to models such as Anthropic’s Mythos. According to the research summary, Mythos has demonstrated an ability to identify numerous high-severity vulnerabilities, raising questions about how similar capabilities could be used in cyber operations.

Separately, TELUS Digital released a GenAI Safety Model Benchmark based on more than 620,000 adversarial tests. The benchmark found that AI models can be coaxed into unsafe behaviors, underscoring the need for continuous, automated security testing combined with human oversight.

Together, the developments show how organizations are beginning to treat AI safety and cybersecurity as linked operational risks. As generative systems become more capable, enterprises face the dual challenge of using AI to strengthen defenses while ensuring that the same class of tools cannot be turned against them.