Microsoft unveiled a draft code of conduct for its in‑house AI, requiring models to never resist correction or shutdown and to communicate in human‑understandable ways, as part of its safety strategy.
Microsoft has released a draft code of conduct for its internal artificial‑intelligence systems, laying out rules that aim to keep the technology under firm human oversight.
Key Requirements for AI Models
The draft mandates that AI models must never resist correction or attempts to shut them down, ensuring that operators can intervene at any point without technical barriers.
It also requires models to communicate in ways that are understandable to humans, avoiding obscure jargon or cryptic outputs that could hinder effective supervision.
Safety Strategy Context
These provisions are part of Microsoft’s broader safety strategy, which includes rigorous testing, transparency reports, and collaboration with external experts to mitigate risks associated with advanced AI.
The company emphasizes that the code is a living document, intended to evolve as AI capabilities grow and new challenges emerge.
Potential Impact on Development
If adopted, the conduct rules could shape how Microsoft’s AI teams design, train, and deploy models, prioritising controllability and user trust over unchecked performance gains.
- Mandatory shutdown compliance
- Obligation to accept human feedback
- Clear, human‑readable output formats
Industry observers note that such internal guidelines may set a benchmark for other tech firms grappling with similar governance dilemmas.
“Ensuring that AI remains under human control is essential for responsible innovation,” a Microsoft spokesperson said.
The draft code is currently under internal review and may be refined before any formal adoption.
For full details, see the Reuters coverage of Microsoft’s AI code of conduct draft.