How Startups Can Help Protect Against AI-Based ThreatsThomvest's Ashish Kakran on How Large Language Models Have Grown the Attack Surface
The enterprise adoption of AI-based large language models has created a new attack surface for adversaries to exploit, said Thomvest Ventures Principal Ashish Kakran.
The performance of large language models depends on the quality of data used to train them, and individuals with access to proprietary data in specific verticals such as healthcare or fintech can add a lot of value. But if a threat actor gains access to or tamper with the data that's been used to train the large language models, Kakran said they could wreak a lot of havoc.
"In addition to the advantages that it gives to your company, it opens up new kinds of attack surfaces," Kakran said. "As you train these models, there are challenges that enterprises have to think about."
In this video interview with Information Security Media Group at RSA Conference 2023, Kakran also discusses:
- How businesses can insert security into the development of machine-learnings models;
- Safeguards to ensure workers don't share critical code with ChatGPT or another LLM;
- How Thomvest will identify the best startups for defending against AI-based threats.
Kakran helps technical founders build lasting companies as a trusted long-term partner. Some of his recent investments include Harness, a DevOps CI/CD automation company, and Isovalent, creator of Cilium and eBPF. Kakran has a deep network of software builders, sales leaders, investors and CXO buyers that he tries to make available to founders.