The Internet Security Research Group report warns of a fundamental technological divide forming in the field of cybersecurity. While large-scale projects may be able to reduce vulnerabilities by 95% using neural networks, Open Source solutions could come under fire due to the emergence of ultra-fast automated attacks.
What Happened
According to analysis by the Internet Security Research Group, using LLMs for code auditing and patching will allow resource-rich projects to reduce the number of vulnerabilities to a 5% level by the end of 2026. Simultaneously, there is a growing risk of so-called N-day attacks, where exploits are created almost instantly after a patch is released thanks to the combination of neural networks and binary diffing.
Context
The development of large language models creates a dualistic environment: on one hand, they radically accelerate the discovery and remediation of bugs; on the other, they scale the capabilities of malicious actors. Access to computing power is becoming a defining factor in defense, leading to a polarization of the software market.
Why It Matters for the Industry
For the industry, this means a necessity for extreme automation of CI/CD cycles. The emergence of tools for instantaneous vulnerability exploitation requires companies to implement LLM auditing directly into pre-commit processes and accelerate the transition to memory-safe programming languages, such as Rust, which neural networks work with more effectively.
Why It Matters for Users
Software security is now directly dependent on computing power budgets. Developers and users of Open Source projects should consider that less well-funded projects may become easy targets for automated attacks, while corporate software will become significantly more resilient.
Sources
Author
Look at AI, Editorial Staff