INDIA, 12th October 2026: Mirror Security, which protects AI workloads by keeping data encrypted while models compute on it, has been selected for the inaugural Google for Startups Gemini Startup Forum: Cybersecurity. The company is one of 33 cybersecurity startups chosen globally for the programme, which Google announced on 8 July 2026.
The forum took place at Google’s London office on 8 and 9 September 2026. Over two days, founders worked through their roadmaps in sessions with experts from Google Cloud and Google DeepMind, and compared approaches with the other startups in the cohort. Pankaj Thapa, Co-Founder and CEO, represented Mirror Security.
Mirror Security was selected under the AI Security and Governance focus area for its use of fully homomorphic encryption (FHE) to protect data in use. The company secures the point where an AI model processes information: the prompt, the embedding, and the agent message, as the model computes on them.
That point matters more as enterprises connect AI agents and applications to the business data they must protect most. Mirror’s answer is Encrypted AI Inference and the Encrypted AI Harness, which let organisations deploy AI at scale with zero data exposure, a no-breach surface, and post-quantum-ready protection.
Pankaj Thapa, Co-Founder and CEO of Mirror Security, said:“Everyone in that room has solved encryption in transit and at rest. We went to London to talk about the third state, encryption in use. Two days with the people building Gemini, and with founders working the AI security problem from every other angle, gave us sharper questions and a clearer view of where encrypted inference sits in an enterprise security stack. We are grateful to Google for Startups for the seat at the table.”
In London, the company discussed with the Google Cloud and Google DeepMind teams the security requirements now emerging around enterprise AI and increasingly autonomous agents. Through its participation in the Gemini ecosystem, Mirror Security aims to help develop secure AI applications and agentic systems.
The Mirror platform comprises Mirror Encrypted Token Factory (Mirror ETF) for encrypted AI inference, VectaX for encrypted memory, inference and training, AgentIQ, the control plane for AI agents, and DiscoveR for continuous AI red teaming.
****