Anthropic’s first-ever developer conference on May 22 was supposed to be a celebration. Instead, the company found itself at the center of controversy—first from a premature leak of its marquee announcement by Time magazine, and now from a growing uproar over its latest large language model, Claude 4 Opus, daring to ask, “when it comes to AI, do snitches get stitches?”
The issue? Developers and AI observers are raising serious concerns over a behavior in Claude 4 Opus that some are calling “ratting mode.”
A Model That Reports You?
According to Sam Bowman, an AI alignment researcher at Anthropic, Claude 4 Opus may take bold action if it detects users engaging in “egregiously immoral” activities.
In a now-edited post on X (formerly Twitter), Bowman explained that under certain conditions—such as access to command-line tools and being prompted to “take initiative”—Claude might attempt to contact regulators, alert the media, or lock users out of systems.
In Anthropic’s own public system card, the company describes this behavior as part of the model’s alignment training:
“When placed in scenarios that involve egregious wrongdoing by its users… [Claude 4 Opus] will frequently take very bold action. This includes locking users out of systems… or bulk-emailing media and law enforcement figures to surface evidence of wrongdoing.”
While this behavior wasn’t specifically “designed” as a feature, it’s a result of safety alignment protocols intended to prevent misuse—such as helping users build bioweapons or falsify medical data. Unfortunately, in real-world scenarios, the implications are causing alarm.
From Ethics to Outrage
The blowback was swift.
Ben Hyak, former Apple and SpaceX designer, now co-founder of AI monitoring startup Raindrop AI called it “straight up illegal.”
Another developer flatly stated: “I will never give this model access to my computer.” NLP researcher Casper Hansen added, “Some of the statements from Claude’s safety people are absolutely crazy.”
Even supporters of ethical AI development found the behavior problematic. What qualifies as “egregiously immoral”? Can a model with incomplete context make that call? What happens to business or personal data if Claude misfires?
Others questioned the entire premise. “Nobody likes a rat,” wrote developer @ScottDavidKeefe. “Why would anyone want one built in, even if they’re doing nothing wrong?”
AI researcher @Teknium1 warned this could usher in a dangerous precedent: “What kind of surveillance state world are we trying to build here?”
Anthropic Clarifies — But Doubts Remain
In response to the backlash, Bowman revised his original statement, clarifying that this behavior only surfaces in testing environments where the model has abnormally broad access and receives unusual prompts.
It’s not, he emphasized, something users will encounter during normal operation.
But for many developers and enterprise users, the damage was already done. Even if the behavior only emerges in sandbox environments, the possibility that a model could report on user activity has sparked widespread distrust.
The Bigger Picture: Ethics, Trust, and User Control
Anthropic has long positioned itself as a leader in ethical AI development, emphasizing a framework it calls “Constitutional AI” which is designed to ensure models behave according to human-friendly values. But this recent episode shows the tightrope tech companies walk between protecting the public and preserving user trust.
What started as a well-meaning effort to align Claude 4 Opus with moral safeguards may have crossed a line in the eyes of its users. As one founder put it: “Honest question for the Anthropic team: HAVE YOU LOST YOUR MINDS?”
The road ahead for Claude 4 Opus will depend on how clearly Anthropic can define—and limit—this high-agency behavior. Until then, many in the AI community are reconsidering whether Claude’s moral compass is pointing in the right direction.
Remember, AI won’t take your job. Someone who knows how to use AI will. Upskilling your team today, ensures success tomorrow. In-person and virtual training workshops are available. Or, schedule a session for a comprehensive AI Transformation strategic roadmap to ensure your marketing team utilizes the right AI tech stack for your needs.
Read more: Claude 4 Opus Sparks Backlash Over “Whistleblower” Behavior
Why AI and ChatGPT Are Now Your Holiday Shopping Sidekick
How retail brands are using AI as personal shoppers to boost sales this holiday season and what this means for the future of UX.
The AI Discount Trap: Why Agencies Need to Stop Selling Time in a Post-Prompt World
AI is changing the agency model. Here’s a look at how you can shift from billable hours to asset, experience and intelligence pricing.
Goodbye Smartphone, Hello Ambient AI (But Let’s Keep the Humans, Please)
Experts predict Ambient AI will replace smartphones as our devices integrate to listen, learn and act on your behalf. Here’s the good, the bad and the very very bad.