Infosys Ltd is pitching its Topaz agent-governance stack to central ministries reviewing artificial-intelligence procurement, aligning the offering with MeitY’s plan to empanel vetted models and expand shared GPU capacity just as Union IT Minister Ashwini Vaishnaw prepares a safety-focused regulation consultation paper.

Public-sector angle

Vaishnaw said at the World Development Report launch on 8 October that government departments would run tenders only against empanelled models or solutions, with startups and large integrators competing on equal footing. Infosys executives told analysts this week that Topaz Fabric and the Responsible AI Office can wrap logging, policy enforcement and human-in-the-loop checkpoints around agents before they touch citizen data.

The company was already the first Indian firm to file under the Hiroshima AI Process Reporting Framework, a voluntary disclosure regime aimed at cross-border trust. Public-sector buyers are now asking whether those governance artifacts can map to India’s forthcoming rules on deepfakes and automated decision-making.

Compute crunch in the background

MeitY’s C-DAC unit issued a request for empanelment to supply up to 10,000 AI compute units, following three bidding rounds that onboarded more than 38,000 GPUs nationwide. Vaishnaw said the next tranche would split capacity between direct government use and commercial bidders.

Infosys does not operate hyperscale GPU farms itself, but Topaz deployments often sit on client-owned or IndiaAI Mission capacity. Ministries piloting document summarization or grievance triage need both compute access and audit trails—pairing MeitY’s hardware push with integrator-side controls.

What buyers should ask

Cert-In’s late-Thursday advisory on AI agents accessing live corporate systems underscores the risk if ministries enable autonomous tools without credential hygiene. Infosys is marketing Topaz Shield components—access monitoring and prompt logging—as mitigations, but contracts will still need clarity on liability when an agent oversteps its sandbox.

Competitors including TCS and Wipro Ltd are pitching similar governance layers after September-quarter earnings calls emphasized “agentic AI” pipelines. The differentiation may hinge on who can certify compliance with MeitY’s empanelment criteria once the consultation paper lands.

Timeline

Vaishnaw asked officials to publish the AI regulation consultation within a month of 8 October, putting ministries on a clock to freeze pilots that lack logging or consent flows. Infosys says several departments are running proof-of-concepts on Topaz; none have announced production rollouts, and MeitY has not published final empanelment rules.

For enterprise technology leaders outside government, the same dynamic applies: GPU access is becoming a commodity faster than governance tooling. Vendors that can document controls may win tenders even if their base models are interchangeable.