Company
Microsoft's in-house AI division, led by Mustafa Suleyman, developing Copilot and other consumer and enterprise AI products and models.
Microsoft AI says its 2.8-trillion-parameter MAI-Base-2 will let it license model weights to trusted partners through Foundry. It also says it tunes its models first for its own products and then for Foundry customers, so licensees would come second.
Reality
Microsoft's MAI-Transcribe-2 covers 60 languages with speaker labels and word timestamps, though streaming is not among its documented features. Live voice products get a fast model for replies and still need another way to hear the caller.
Perspective Coverage
8 publishersReality
David AI's 21-language benchmark found speech systems' error rates spread an average of 25.4 points in five Indic languages, against 3.7 in the tightest five. In those languages, choosing a system is a 25-point decision that an English score can make look like a close call.
Reality
The 37-page draft bars help with mass-harm weapons, cyberattacks and nonconsensual deepfakes, and sets shutdown rules Microsoft says its models cannot be configured around. Comment closes in late October and the principles guide 2027 releases.
Perspective Coverage
8 publishersReality
Microsoft's draft Humanist AI Code of Conduct blocks its MAI models from producing working exploit code, attack tooling and evasion techniques, and the firms deploying those models cannot switch the block off. Comment closes in six weeks.
Reality
Mustafa Suleyman's essay accuses Anthropic of training Claude to treat its own consciousness as an open question. It landed two days after his own unit put a rival code of conduct out for consultation.
Perspective Coverage
3 publishersReality
Mustafa Suleyman's essay and BBC interview criticise how Anthropic trains Claude to act human. The four capabilities he warns about are granted separately, by different people, in every agent rollout.
Perspective Coverage
7 publishersReality
Warnings that AI is about to escape human control rest on the forecasts of well-known figures. New Scientist put the question to two researchers, who point to cyber-capability tests that their own designers did not isolate.
Reality
Anthropic says Claude led about 26% of its R&D in August and wrote more than 80% of the code merged into its codebase, and a thin prediction market gives its next Opus an 83% chance of shipping by September 24.
Reality
Mustafa Suleyman went on CNBC on September 18, days after Microsoft published its Humanist AI Code of Conduct, and argued that Beijing's own May 2026 guidelines on agentic AI undercut the case for waiting.
Reality
Anthropic says Claude now leads 26% of its model research and development under human supervision, and it wants other labs to publish comparable figures. The labs count different things.
Perspective Coverage
4 publishersReality
Microsoft published the draft on Monday after five to six months of work. It applies only to the company's own MAI models, and no model is trained on it until the six-week public consultation closes.
Perspective Coverage
4 publishersReality
Trump says the only guardrail AI needs is a strong and smart president. The controls a product team can actually cite are Anthropic's own plug-pulling, Microsoft AI's new outline and Monday's market selloff.
Reality
The draft code of conduct Microsoft AI published on September 14 is not training anything yet, and a revised version lands toward the end of the year to guide model development from 2027, so the text is still open to argument.
Perspective Coverage
3 publishersReality
Microsoft's new code of conduct binds only the MAI models it builds in house, and it rejects the model welfare research at Anthropic, where Microsoft has $5bn invested and whose Claude models it sells inside Copilot.
Reality
The code rejects model welfare, personhood and rights, and it commits Microsoft's own models to failing a task before breaking its rules. Anyone drafting an internal AI standard now has a vendor text to copy from.
Reality
The early-bird rate on MAI-Transcribe-2 sits 72% below the line's April price, which saves a 100,000-hour buyer $26,000 a year and takes 26 cents out of every 36 a standalone vendor charged for the same hour.
Reality
Luke Metz joins Superintelligence Labs after a three-lab loop. The $14.3bn Scale stake and billion-dollar packages set a price rivals now have to match from capital rather than revenue.
Reality