Google DeepMind uses a Frontier Safety Framework to manage extreme risks in AI models. Manish Gupta said, "We treat safety not as an afterthought, but as a core scientific discipline." The team uses internal red-teaming and external experts to find vulnerabilities. These steps ensure that new technology stays safe for everyone.
At Google DeepMind, we treat safety not as an afterthought, but as a core scientific discipline built into foundational architectures. Guided by our Frontier Safety Framework, a set of protocols designed to prepare for and monitor extreme risks, our models undergo rigorous evaluations throughout their development lifecycle. This includes internal red-teaming to systematically uncover vulnerabilities such as prompt injection, alongside valuable external evaluations from third-party experts to ensure safety before public release.