What is Rubrik?
Rubrik is a data security and cyber recovery company whose engineering leadership has described building agentic, AI-assisted workflows into its products. The material tracks how Rubrik’s technical staff frame the division of labor between large language models and deterministic recovery logic, and how they push agent-first thinking across product and engineering teams.
Company timeline
- Jun 2026 - Devvret Rishi said Rubrik had processed trillions of tokens inside Sage without observing incidents or evidence that the approach could be circumvented.
- Jun 2026 - Rishi described post-training small language models for a use case, shipping the same model across customers and customizing at inference time per customer.
- Sep 2026 - Abhijit Sharma said Rubrik uses LLMs to help generate plans, while the recovery plans themselves in a cyber scenario remain deterministic and explainable to the customer.
- Sep 2026 - Sharma said the goal is to democratize agent-first thinking so every PM and engineering team builds and optimizes use cases rather than a central team.
- Sep 2026 - Sharma said people will make more mistakes than agents if agents are built with the right guardrails.
Where it appears in the record
Every line below is attributed to a named speaker.
Rubrik uses LLMs only to generate cyber-recovery plans, keeping plan execution deterministic and explainable rather than probabilistic end-to-end.
“Instead, what we're using is we're using like all the LLM to help come up with the plans, but the plans themselves of how you're going to do a recovery in a cyber scenario are deterministic and then are explainable to the customer.”Abhijit Sharma · 2 Sep 2026
Rubrik deliberately avoids a centralized AI team, instead requiring every PM and engineering team to build with an agent-first mindset, pushing against the common 'center of excellence' model.
“How do we democratize this so that we don't have a central team that is building all of the use cases and optimizing all of the use cases, but every PM and every engineering team within our company is thinking from an agent first mindset and in agentic workflows.”Abhijit Sharma · 2 Sep 2026
Rubrik Sage, a post-trained SLM shipped uniformly across enterprise customers with inference-time customization per org, worth watching as a deployment pattern for scalable AI security.
“Post-trained SLMs for use case, yes. Ship the same one across customers and do an inference time customization for each customer.”Devvret Rishi · 16 Jun 2026
Rubrik's Sage system has processed trillions of tokens with zero observed circumvention incidents.
“We haven't seen like we process trillions of tokens, you know, inside of Sage, and we haven't seen any like incidents or evidence where that kind of can be circumvented.”Devvret Rishi · 16 Jun 2026
Abhijit Sharma on agents versus humans in complex enterprise setup.
“People are going to make mistakes more than agents if we build our agents with the right guard rails around it.”Abhijit Sharma · 2 Sep 2026