🧠 Food for thought

Implications, context, and why it matters.

Securing the code beneath AI systems

• “Software flaws” means weaknesses in the code that supports AI systems, not only model issues such as prompt injection 1.
• TrendAI said its AESIR research platform checks whether a flaw can be reached and exploited. It found a remote code execution bug with a CVSS score of 9.8 in Nvidia’s Isaac GR00T robotics platform, tied to a deserialization error in the “TorchSerializer” class 1.
• TrendAI’s analysis says AI Common Vulnerabilities and Exposures (CVEs), the public catalog of disclosed software flaws, topped 1,000 in 2025. That was 70% higher than a year earlier 1.
• Trend Micro said it worked with the Zero Day Initiative, its program for finding previously unknown flaws, on patches for AI platform bugs in products from Nvidia and Tencent 2.

AI security is breaking into specialties

• TrendAI describes the market as two related areas. One covers model-level risks, and the other covers the software infrastructure beneath AI 1.
• Anthropic, the company behind Claude, provides Claude Opus 4.7 for TrendAI’s security research and its AESIR platform 2.
• That setup supports a business model where AI helps find vulnerabilities at machine speed, with human experts reviewing the work before it feeds a defensive platform 3.
• Enterprise spending on generative AI rose from US$11.5 billion in 2024 to US$37 billion in 2025, a 3.2x jump. TrendAI says that growth makes the software stack and supply chain harder to secure for businesses using these tools 1.