Trusting AI with Code: How Secure is AI-written code?
summary
97% of developers use AI code helpers and experiments find about 40% of generated apps contain known vulnerabilities, with 20% of orgs reporting serious incidents and 69% of CSOs seeing AI-linked problems. GPT-5 narrows the gap between default and “make it secure” prompts but hallucinated or out‑of‑date dependencies (training cutoff Aug 2025) and supply‑chain malware explode risk — one NPM compromise reaches massive exposure in 4.5 hours. An open‑source scanner called Intel built on open grep is used to triage and classify likely malware.
AI-written summary · about
chapters (9)
- 0:00AI code security intro
- 1:10Real incidents with AI code
- 2:13Testing AI code security
- 3:36Secure prompts impact
- 4:39AI and software supply risks
- 6:27Malware speed in AI chains
- 9:17AI for non-deterministic threats
- 10:43Rethinking security tools
- 11:26Closing remarks
related talks
55mUsing GenAI on and inside your code, what could possibly go wrong?
52mUsing GenAI on and inside your code, what could possibly go wrong?
23mVibes Not Vulns: Securing the Era of AI-Written Software
31mHacking AI Applications
26mSmarter Code, Sneakier Risks: Supply Chain Security in the Age of AI
34mTrust in the Age of AI Agents | Building Trustworthy AI
59mUsing AI for Secure Code Creation: Enhancing Software Security
01h 24mOffensive Cybersecurity by AI Promises and Pitfalls