Microsoft and 1Password highlight AI challenges in software security

At Black Hat USA, David Weston, vice‑president of AI security at Microsoft Corp., warned that artificial intelligence is making offensive cyber capabilities cheap and abundant, ending the era of rare zero‑day exploits. He urged security teams to shift from reactive patching to building software that is secure by construction, emphasizing memory‑safe languages such as Rust and formal verification methods.

A separate study by Off‑By‑1 Labs, the security research arm of 1Password, tested six recent open‑source vulnerabilities and found that AI‑generated patches were usable in only 26 % of cases, with the remaining 74 % containing defects. The research concluded that current large language models are not yet reliable for automatically fixing security flaws, underscoring the need for human‑driven, safety‑first development practices.