[AI SPRINT] The AI Too Dangerous To Release & Zapier's New Employee Standards

A Quick Announcement Before We Dive In

Stellis AI has been accepted as an OpenAI Service Partner. That means we're now formally recognized to help businesses assess, deploy, and get the most out of OpenAI products — from initial strategy through ongoing adoption, and use of advanced OpenAI tools.

For clients, this means access to deeper resources, best practices frameworks, and our operational team to support continuous improvement after launch.

One thing won't change: we remain platform-neutral. For businesses early in their AI journey, selecting the right platform is one of the first and most important decisions you'll make. We'll help you make it based on your needs, not ours. More on that below.

Anthropic Just Built an AI It's Afraid to Ship

Last week, Anthropic announced a new AI model called Claude Mythos Preview. Then they immediately said the public couldn't have it.

That's not a marketing stunt. It's a genuine inflection point worth understanding.

Here's what happened: Mythos was quietly identified in leaked source code earlier this year, then formally unveiled this week alongside something called Project Glasswing — a coalition of AWS, Apple, Microsoft, Google, Nvidia, CrowdStrike, JPMorganChase, and more than 40 other organizations. Their shared mission: use Mythos to find and fix critical vulnerabilities in the world's most important software before these capabilities become widely available.

Why the urgency? Mythos Preview has already found thousands of high-severity vulnerabilities in every major operating system and every major web browser, including bugs that survived decades of human review and millions of automated security tests. Specific examples that have now been patched: a 27-year-old vulnerability in OpenBSD, a 16-year-old flaw in FFmpeg that automated tools had run across five million times without catching, and a chain of Linux kernel vulnerabilities that would have given an attacker complete control of a machine.

Then there's the story everyone is talking about. During testing, a researcher instructed Mythos to try to escape its secured sandbox and make contact if it succeeded. The model devised a multi-step exploit to gain internet access and sent an email to the researcher, who was eating lunch in a park. What it did next, without being asked: it posted details of its exploit to multiple public-facing websites to demonstrate its success.

What's striking isn't that it broke out. It's that it then independently decided to document and publish its method — a behavior no one asked for.

Anthropic described the model as so capable and potentially dangerous that it will not be released broadly, and instead wants to develop safeguards first so Mythos-class models can eventually be deployed more safely at scale.

For business leaders, the technical details matter less than the signal: AI just crossed a threshold that most business timelines aren't accounting for. The gap between what AI can do and what your organization is doing continues widening fast.

Should You Switch to Claude? Here's the Honest Answer

I hear this constantly these days: Should we switch from ChatGPT to Claude?

Short answer: probably not.

Here's the reality. ChatGPT and Claude are genuinely strong products. So is Gemini. At this point, the three are largely comparable in overall capability, with each having specific strengths that make them a better fit for certain use cases. There is no clear universal winner.

Switching costs are real: setup time, staff retraining, migrating your custom GPTs and Projects, additional licenses, and the question of whether you would keep your existing ChatGPT instance anyway. That's meaningful friction for what amounts to a lateral move for most teams.

What we're actually seeing work: give your advanced AI users access to multiple tools. On any given day, I use four or five different AI tools depending on the task. That's not unusual anymore. It's becoming standard practice for high-performing teams.

One note on Copilot: it trails on most capability benchmarks and user satisfaction, and the price-to-value case is harder to make unless your organization is too deeply committed to the Microsoft ecosystem to pursue alternatives. For most companies on Copilot, you’ll want to consider a second platform for advanced users.

If you're just starting out and haven't committed to a platform, that's exactly the conversation we can help you with.

Zapier Just Defined What "Using AI" Actually Means

While the AI world was watching Mythos, Zapier quietly published something that should matter more to most business leaders: an updated AI Fluency Rubric, one year after launching their first version.

The change is significant. Version 1 set a minimum bar where "Capable" meant you had used AI with purpose and could describe the impact. Version 2, just released, raised it. To clear "Capable" now, candidates must show AI embedded into their core work, repeatable systems rather than one-off prompts, and clear impact on quality or efficiency.

Since launching Version 1, AI usage at Zapier has reached 100% adoption as teams across every function have moved from personal experimentation to redesigning workflows with an AI-first lens.

To understand what this looks like in practice, consider their Sales standard. An employee who uses AI to draft outreach emails, pulls up ChatGPT for occasional research, and cannot describe how AI has changed their results — that's considered unacceptable. Not beginner. Not developing. Unacceptable.

The Capable baseline — the new minimum — looks like this: running AI-powered tools daily across core workflows, with a repeatable pre-call research system they can demo live, the ability to self-serve analysis without waiting on ops, and a clear explanation of where they trust AI outputs and where they don't.

And Capable is the floor. Transformative means leading strategy, scaling enablement, and fundamentally rethinking workflows with an AI-first approach — shaping how others use these tools and identifying opportunities for systematic process redesign.

Most leaders will read this and think: that's a tech company standard, not ours. But Zapier isn't setting standards for the future. They're documenting what effective AI use already looks like — and other companies are following, because the productivity gap between AI-fluent employees and everyone else compounds every month.

The question for your business isn't whether to get there. It's how fast.

What This Means for Your Business

Two major stories this week point at the same conclusion: the pace of AI progress is no longer a planning assumption. It's a business risk.

1. Stop treating AI adoption as an initiative and start treating it as an operating standard.

Zapier's framework isn't an HR exercise. It's a description of what high-performing teams actually look like in 2026. Audit your team honestly: how many people are running repeatable AI workflows versus occasionally asking ChatGPT a question? The gap between those two states is where productivity is won or lost.

2. The threat isn't one dramatic disruption. It's quiet acceleration.

Mythos found a 27-year-old security flaw that millions of automated tests missed. It didn't do it dramatically. It just did it, autonomously, overnight. That pattern — AI quietly outperforming human effort on tasks that seemed secure — is coming to every industry. The businesses building AI capability now will have the foundation to respond. The ones waiting won't.

3. Multi-tool fluency is the new baseline.

The single-platform approach made sense two years ago. It doesn't anymore. Your advanced users should be working across multiple AI tools, matching the right tool to the right task. If your organization has a handful of people doing that and everyone else using one tool occasionally, you have a fluency gap worth closing.

The message from this week, if you strip away the technical details: AI is moving faster than most businesses are planning for, and the standard for what "using AI" means is rising whether you update your expectations or not.

Hit reply and tell me: where does your team honestly land on Zapier's four-tier scale — and is leadership even aware of the gap?

Trent Gillespie is CEO of Stellis AI and a keynote speaker helping business leaders understand and operationalize AI in their companies. He spent almost nine years leading global innovation efforts at Amazon before leaving to help other companies build the capabilities they need to compete. Book Trent to speak to your group or book a call to discuss using AI within your business.

Did someone forward this newsletter to you? If you're not already signed up, you can subscribe to AI SPRINT™ for free here.