The Biggest AI Labs Are Starting to Coordinate on Safety

Morning, {{first name | folks}}! Today, we’re looking at AI safety talks between the biggest labs, Microsoft’s new AI code of conduct, Meta’s expanding paid AI offering, Factory’s $5 billion funding round, and Autodesk’s new AI tools for engineering.

OpenAI, Anthropic, and Google DeepMind have been talking about AI safety for the past few weeks, according to OpenAI’s global policy chief Chris Lehane. The conversations are focused on risks that come with increasingly capable AI systems, while OpenAI is also backing bipartisan proposals in the US aimed at addressing catastrophic AI risks.

That puts the companies in an interesting position. They’re competing to build more capable models while also trying to work out where safety needs to keep pace. Meanwhile, President Trump has pushed back against calls for stronger AI guardrails, calling concerns about AI risks a “hoax.”

Microsoft AI published its first Humanist AI Code of Conduct today, open for public feedback for six weeks before it shapes real model training in 2027. The core stance stands out hard against the rest of the industry, Microsoft explicitly rejects building "an unbounded and unlimited entity with high degrees of autonomy," aiming instead for AI that stays domain-specific and bounded, a pointed contrast to the recursive self-improvement warnings you've seen from OpenAI and Anthropic's own leadership this month.

It also draws a firm line on AI personhood: no rights, no welfare claims, no pretending to have feelings, and models are required to actively discourage users from forming emotional attachments to them. Absolute limits apply no matter what an operator or user wants, no weapons help, no offensive hacking, no evading shutdown. It's still just a draft though, current Microsoft models aren't trained on any of this yet.

Meta is launching Meta One, a new subscription service that brings more than 50 features across Facebook, Instagram, WhatsApp, and Meta AI. The plans include expanded AI image and video tools, audience insights, analytics, scheduling, team collaboration, and other features aimed at creators and businesses. Meta says its existing paid plans have already reached 15 million subscriptions and trials.

Meta One also expands the amount of AI access available to paying users, alongside tools for managing content, audiences, and business activity. The company is positioning the subscription as a broader layer across its apps rather than a single premium feature, bringing more of its AI and business tools into one paid service.

Factory has raised $200 million at a $5 billion valuation, more than tripling its valuation from $1.5 billion in April. The round includes Blackstone, Khosla Ventures, Sequoia Capital, and Insight Partners. Factory builds AI agents for software development, covering tasks such as coding, testing, and maintaining code.

The company says it now has hundreds of thousands of developers using its platform and plans to use the new funding to expand its research and product development. For developers, the shift is moving beyond AI that helps write code toward agents that can take on larger parts of the development process, including testing, debugging, and maintaining software.

Autodesk is adding new AI features across Fusion, Inventor and Vault, including tools that can automate parts of design, assembly, and product data management. Fusion is getting AutoTimeline and AutoAssemble, while Autodesk is also introducing an agentic PLM experience that can work across product information and engineering tasks.

As these tools move from assisting with individual tasks to handling parts of the work themselves, engineers can spend less time on repetitive steps and more time reviewing designs, making decisions, and solving problems that need human judgment.