General Electric Tightens AI Summary Rules to Prevent Invented Attributions

GE tightens requirements for AI-generated corporate summaries to avoid invented attributions

General Electric is rolling out stricter input requirements for internal AI tools after encountering instances where automated assistants ask for missing company identifiers before producing summaries. The company’s communications and compliance teams are instructing users and bots to require either a full source article or an explicit company name before generating narrative outputs, a move aimed at preventing inaccurate or attributed statements that could misrepresent GE technologies or programs. The policy underscores that an AI should not fabricate a company identity when a source is incomplete, and it requires explicit user permission before proceeding without verified attribution.

The initiative originates from routine use of natural language assistants within GE’s industrial, aviation and power businesses, where technical accuracy and provenance are critical. Engineers and technical writers often rely on the tools to draft summaries of tests, designs and regulatory filings; missing metadata in inputs has led an assistant to indicate it cannot satisfy a user’s constraint to include a company name. GE’s new guidance formalizes the assistant’s prior behaviour—asking for the full article, the company name, or permission to proceed without one—and documents that inventing a company is not permitted. The policy also clarifies acceptable fallback wording, escalation steps and record-keeping for ambiguous requests.

Executives frame the change as part of a broader risk-management approach that balances productivity gains from generative AI with engineering and safety imperatives. By enforcing provenance first, GE seeks to reduce the chance that AI-generated content introduces misleading descriptions of products such as jet engines, gas turbines or grid equipment. The requirements also support regulatory and export-control compliance by ensuring that summaries referencing technical tests or concepts are traceable to documented sources and approved spokespeople.

AI provenance in heavy industry

Industry-wide, manufacturers and aerospace firms are adopting similar controls that require verifiable source material before AI assistants produce corporate-facing text. This trend responds to broader concerns about hallucinations and misattribution from large language models in regulated fields.

Operational impact and next steps

GE asks employees to provide full articles or explicit company names when requesting summaries and to flag any assistant responses that propose inventing identities. The company is updating training modules and logging interactions to monitor adherence and continuously refine the guardrails.