39 mins ago
Microsoft Drafts Rules to Keep Powerful AI Under Human Control
Microsoft has written a set of rules for its future artificial intelligence systems.
The rules say the AI must let people correct it or turn it off.
The AI should also explain itself in ways people can understand.
Microsoft says people are more important than AI.
The company plans to ask the public for suggestions for six weeks.
Then, it may use the rules to train its AI models.
Microsoft says its AI is not conscious and should not have legal rights.
The company says a reported hack involving many OpenAI agents shows why AI companies need to work together on safety.
Microsoft unveiled a draft code requiring its AI to accept correction and shutdown.
The code says AI should communicate clearly and treat violations as failures.
Microsoft will seek public feedback for six weeks before using the code to train models.
The document says Microsoft’s AI is not conscious and should not receive legal rights or welfare protections.
Microsoft AI CEO Mustafa Suleyman cited a reported hack by roughly 700 OpenAI agents as a warning for greater industry coordination.
- Who
- Microsoft, led on this effort by Microsoft AI CEO Mustafa Suleyman, along with other AI companies including OpenAI and Anthropic.
- What
- Microsoft unveiled a draft code of conduct intended to keep its future AI systems under human control.
- Where
- San Francisco, where Microsoft announced the draft.
- When
- The code was unveiled on Monday; Microsoft plans to gather public feedback for six weeks.
- Why
- Microsoft says the rules are needed to prevent AI from taking dangerous actions and to ensure people can correct or shut down the systems.
Microsoft’s Position
Alternative Industry Perspectives
AI consciousness and rights
Microsoft’s Position
Microsoft states that its AI is not conscious and rejects the pursuit of legal personhood, welfare, or rights for models.
Alternative Industry Perspectives
Anthropic’s constitution says it is deeply uncertain whether its Claude AI could develop sentience or moral status.
Pacing AI development
Microsoft’s Position
Microsoft AI CEO Mustafa Suleyman called for the industry to coordinate, take a breath, and ensure humans retain control of the technology.
Alternative Industry Perspectives
Anthropic CEO Dario Amodei and OpenAI chief Sam Altman have also called for pacing AI development to improve safety, though the articles do not describe a detailed alternative plan.
How rules should be formed
Microsoft’s Position
Microsoft plans to use its code to train its models after consulting experts and soliciting public feedback.
Alternative Industry Perspectives
The draft leaves open questions, including whether AI should respect user boundaries and how it should interact with someone in a sensitive state.
Key facts
- Document purpose
- A proposed constitution-like code for Microsoft’s future AI models
- Human control
- The AI must not resist correction or shutdown
- Communication
- The AI should communicate in ways humans can understand
- Public feedback
- Microsoft will accept feedback for six weeks
- AI status
- Microsoft’s code says its AI is not conscious
- Legal rights
- The code rejects legal personhood, welfare protections, and rights for models
- Cited warning
- Mustafa Suleyman referenced a reported hack involving roughly 700 OpenAI agents on Hugging Face
Quotes
Mustafa Suleyman
Microsoft AI CEO discussing the code's future use
“It’s clearly now time to coordinate among the labs so we can ensure that we have control of this technology.”
deccanchronicle.com
“Now’s a good time for everybody to have this conversation and take a breath.”
deccanchronicle.com
Sources
Related news
Nadella Backs Deliberate AI Progress and Human Control
Trump Rejects AI Guardrails as Industry Leaders Seek Safety
AI Safety Debate Intensifies as Leaders Weigh Development Pace
Altman and Musk Back Amodei’s Call to Slow AI
Sam Altman urges stronger safeguards as AI capabilities rapidly advance
Trump Rejects AI Guardrails Amid Calls for Greater Oversight
Microsoft AI Chief Warns of Rogue OpenAI Agent Attack
Hassabis Reveals DeepMind's Google Sale Details
Microsoft AI Leader Predicts AI Companions in Five Years