The Ministry of Electronics and Information Technology (MeitY) has advised government ministries not to deploy artificial intelligence models from OpenAI and Anthropic for cybersecurity and related functions for now, according to a report by ThePrint. The move comes days after representatives of both companies met officials across multiple ministries to propose the use of their AI models for cybersecurity and similar government functions.

According to the report, an office memorandum circulated by the MeitY department asks ministries not to put such models into use prematurely. While MediaNama could not independently confirm the exact number of ministries approached or the level of the meetings, the report says officials from both companies met with multiple government departments.

Wider curbs on external AI tools: The latest direction follows a broader push by government agencies to limit the use of external AI tools for official government work. On June 29, Moneycontrol reported that several government organisations had issued internal directions asking employees not to use unapproved external AI platforms to process or share official, confidential or sensitive information.

One of the internal communications reviewed by the publication stated: “Restriction on use of unapproved external AI platforms/tools for processing or sharing official/confidential/sensitive data..”

Government’s evolving position: The latest advisory reflects the government’s evolving approach to AI use in official work. In January 2025, the Finance Ministry asked employees to avoid using AI tools such as ChatGPT and DeepSeek on office devices, saying they posed risks to the confidentiality of government data and documents. Two months later, however, the Centre told Parliament that there was no blanket ban on AI tools in government offices. It said officials could use AI while ensuring the security and confidentiality of government data and following existing cybersecurity rules. In November 2025, the government released the India AI Governance Guidelines, which adopted a risk-based approach to AI. The guidelines said high-risk AI systems should not be deployed without safeguards but did not create new legal restrictions, instead relying on existing laws and sectoral regulators.

The June advisory did not prohibit the use of generative AI tools altogether. Instead, it directed government organisations to use only approved AI platforms for sensitive work while strengthening cybersecurity practices.

CERT-In flags cyber risks: The advisory followed a recent warning from the Indian Computer Emergency Response Team (CERT-In), which cautioned government organisations that advanced AI models could enable more sophisticated cyberattacks and significantly reduce the time available to fix security vulnerabilities. CERT-In advised organisations to patch critical vulnerabilities within 12 to 24 hours.

Along with restricting unapproved AI tools, the guidance asked organisations to strengthen multi-factor authentication, install security updates on time, conduct regular vulnerability assessments and security audits, monitor internet-facing systems, maintain secure offline backups, report cybersecurity incidents to CERT-In, and appoint nodal officers responsible for cybersecurity preparedness and compliance.

Shadow AI concerns: The government’s cautious approach may also reflect concerns around “shadow AI,” the use of unapproved AI tools by employees without organisational oversight. Such use can expose confidential documents or sensitive government information when they are uploaded to public AI platforms, while making it harder for departments to monitor data flows, investigate security incidents and enforce cybersecurity policies.

Taken together, the June guidance and the latest MeitY communication suggest that the government is tightening oversight of frontier AI models in official systems, particularly for cybersecurity, while assessing the risks of deploying external AI services in government operations.

Source link