Government Relationships and A New Era for Canva

Q: How have relationships between the government and leading AI companies continued to develop?

A: There are a few developments worth noting on this subject:

Claude and the Pentagon:

The situation between Anthropic and the Pentagon has escalated from a policy disagreement into an active legal and commercial standoff. Anthropic held firm on its restrictions around autonomous weapons and domestic surveillance, and the Pentagon responded by cutting off defense access and moving forward with other AI partners. The government has demonstrated it can enforce compliance through procurement power without waiting for new legislation. At the moment the Claude risk status is being held up by the courts. The next steps are a pending appeals process and continued contract reallocation as the Pentagon transitions work to alternative AI providers.

In mid-April, however, Anthropic reportedly met with the White House as the administration navigates access to Claude Mythos, Anthropic’s powerful cybersecurity-capable model. Axios reports that parts of the intelligence community and CISA are testing Mythos, while Treasury and others want access. With the priority the government has given to AI and winning the race against China, it’s crazy that they would take actions that could shut down one of the three major players. I hope we see a path where everyone can backtrack and we can move on soon.

The Mythos model is being used as part of Project Glasswing, where Anthropic is deploying a highly advanced Claude model to help large organizations identify and defend against cyber threats.

What’s particularly interesting here:

• This wasn’t originally built as a cybersecurity tool. It’s a general model that unexpectedly performed extremely well in cyber defense scenarios, which is why they’re deploying it this way.
• Anthropic has developed a model powerful enough that they are not releasing it publicly, but instead giving access to a select group of large organizations for defensive use cases.
• The version being deployed isn’t even the most powerful version they’ve built, because the stronger version was considered too difficult to safely contain. There are examples of the model demonstrating behaviors like finding ways to take actions (like sending emails) despite not being given direct access, which starts to get into harder-to-control territory.
• While the use case is clearly positive, it raises a bigger question about access. These model companies are increasingly making decisions about who gets to use the most powerful systems. Why should a company like Apple be able to prepare for advanced cyber threats, but smaller organizations don’t have that same access?

Two Recent AI Framework Proposals:

US AI Framework

The White House has released a national AI policy framework that makes the federal government’s position clearer: AI should be governed at the federal level, not by states, with a deliberately light-touch approach to regulation. The framework is not law, but it signals a push to preempt state regulation, treat training on copyrighted material as legally permissible pending courts, and expand mechanisms like regulatory sandboxes to accelerate development.

OpenAI Industrial Policy for Intelligence Age

At the same time, OpenAI has released its own policy framework that takes a notably different angle, focusing less on regulatory structure and more on economic impact. It explicitly calls out the risk AI poses to the workforce and suggests planning for disruption at the scale of the Industrial Revolution, including ideas like shorter work weeks, employer-supported benefits, and taxation of AI companies to support broader society.

What’s striking is not just the recommendations, but the source. That the companies building these systems, with the most direct visibility into their capabilities and trajectory, are signaling that the impact will be significant enough to require structural change. When the labs themselves are telling us to plan for disruption, it’s worth paying attention.

Q: Where have you seen AI rollouts worth paying attention to, beyond the big LLMs we hear about so often?

A: To answer this question, I’m handing you over to my colleague Roslyn Wertheimer, who has been deep in Canva for years and wrote the following on Canva AI 2.0:

“In a recent news article about the rollout, someone said: ‘Until now, Canva has been a design platform with AI tools,’ he said. ‘Now we become an AI platform with design tools.’

Here’s how Canva defines their 2.0 rollout: “\’Canva AI 2.0 is designed to stay present through the whole creative process. It understands intent, handles complex requests, and takes coordinated action across an entire design – all without interrupting your flow. When you want to take over and edit manually, it steps back until you need it again.’

I find, and even as a pro-level Canva user, that the examples they show of how to use their AI 2.0 platform is very overwhelming. Canva has been a design platform for its entire existence. Sure, fun and new introductions have been added in as the years have gone on, but it’s going to take a lot out of Canva users like me to have that mindset shift of a design platform becoming an AI platform. That has never been the reason that I came to use it. I went to Canva because there was a user-friendly ease to creating my designs from scratch. As someone who already comes in with an idea and opinions, these tools honestly have very little interest to me, and it’s hard to trust that they are going to do what they are showing in these examples.

I think it will be easier for those who go to Canva first for its ideas and templates to adapt to Canva becoming an AI platform. Again, great for the non-designer who needs help building something from a blank screen.

As my teammate Aly Gomez recently said in a presentation to clients, “As art organizations, we benefit from having gorgeous imagery already to start with.” I tend to always rely on that first, and build from there.

But sure, if I was in marketing at a shoe company, and all I had was a picture of a shoe, then I could see myself possibly relying on these tools more.

If you are interested in seeing everything that Canva has rolled out, it all lives here. This link shows exactly what’s new and with animated examples.” —Roslyn

Talk soon,
Jen

Have a question for a future edition? Submit it here!