Your AI coworker is taking all the credit

When Aubrey wrapped up a big work project earlier this year, her manager emailed her with an unusual request. In her presentation to the senior leadership, could she highlight how Claude, the AI chatbot, helped her?

Aubrey, a New York-based healthcare analyst, spent countless hours working on a new way to speed up an expensive medical manufacturing process. While she used Claude in a small capacity, her manager wanted her to make it sound as if the AI had come up with the idea and executed it on its own.

“I had worked for over a year to gather issues, draft alternatives, learn the implications of any changes, and my manager wanted to credit all of that to AI,” says Aubrey, who requested to omit her last name for fear of retaliation. She chose the middle ground: in her planned presentation, she puffed up AI’s role, while still communicating that she did much of the heavy lifting. But in the middle of her presentation, her manager interrupted her, announcing that she had built it all out in a minute with AI. Weeks later, Aubrey received a less-than-enthusiastic annual review. While her boss didn’t directly mention the incident during the review, he did say it was a factor when she asked later.

Deepak, an India-based IT developer for a Fortune 500 tech company, recently found himself in a similar boat. Over a year ago, he started regularly crediting the automated coding agents he deploys to carry out the grunt work for transparency, but soon, he says, his workplace’s upper management began to assume all his positive contributions came from AI, and he suspects it has stalled an expected promotion.

Deepak and Aubrey aren’t alone. White-collar workers are caught in a nasty pickle. If they acquiesce to their bosses’ demands to use AI more, will their bosses believe the AI is doing the work for them — are they building their own career guillotines? As a result, many employees have begun hiding their AI usage and wonder how much credit, if any, they should give it for their efforts.

In an era of mass AI-driven layoffs, to credit or not to credit AI can feel like as vital a question as to be or not to be.

Christoph Riedl, an information management professor at Northeastern University, says people’s hesitation to disclose AI assistance is entirely justified. In a recent meta-analysis, Riedl and his coauthors examined 13 studies spanning a variety of job functions and titles to assess how managers treated their employees after employees disclosed their use of AI. The conclusion was clear: managers consistently devalued workers’ contributions to projects when workers revealed AI had assisted them; managers assumed the technology did most of the heavy lifting. One of the few ways for people to avoid this “AI penalty” was to retain agency over their core work and to outline precisely how they contributed to a task. That may be easier said than done, however, as employers adopt methods to track AI use that can obfuscate the extent to which humans retain influence over their creative output.

Most companies now rely on tracking tokens, the fundamental unit of data processed by an AI model. Seeing how many tokens an employee uses allows managers to see how often they queried the chatbot, the volume of information exchanged, and the length of each interaction. It doesn’t, however, offer insights into what the AI contributed creatively to the project. Anyone could therefore ask their company-issued chatbots endless irrelevant questions about the weather or their personal lives and still appear to be a pro AI user. It hasn’t taken companies long to realize it can be counter-productive and discourage frivolous tokenmaxxing. Last month, Amazon shut down an internal leaderboard that tracked AI token use, as it pushed staff to perform tasks that didn’t necessarily solve any problems.

“Please don’t use AI just for the sake of using AI,” Dave Treadwell, an Amazon senior vice president, told staff at a companywide meeting.

Even more sophisticated methods can be troublesome. AI coding assistants like Claude Code go as far as to automatically add a co-authorship signature in the code they write, without explicitly pointing out which lines were auto-generated or how extensively the human author was involved.

“Our analysis shows that if AI use is disclosed without specific details about how it was used,” Riedl tells me, “the manager’s default assumption seems to be that it was used in a way that reduces agency.” In other words, bosses assume the bot must be the driving force behind that new product feature, the quick software fix, or the text in a lengthy report. “So the detail of how AI was used seems to matter enormously.”

There are some attempts to better understand the exact balance between human and AI contributions. Graham Neubig, a computer science professor at Carnegie Mellon University, cofounded OpenHands, an open-source AI coding platform that adds a sort of footnote-like attribution to a line of code generated by AI. Neubig felt it was important to tag the code as AI-generated to moderate the level of trust and increase the scrutiny a reviewer may place on it.

A team at IBM created an even more detailed way of tracking contributions. The AI Attribution Toolkit was inspired by CRediT, the Contributor Role Taxonomy, a standardized system scientists use to outline each author’s precise contributions in a published paper. On the AI Attribution Toolkit form, people can punch in how much of the work was auto-generated, whether the chatbot produced the content from scratch, and whether certain elements were human-reviewed. The tool, then, produces an attribution statement people can add to their documents, code, and more.

Jessica He, one of the toolkit’s designers, says high-level acknowledgments of AI use are insufficient for both people consuming AI-assisted content and AI users. The way people engage with someone’s work can differ depending on whether AI was used to generate new ideas or simply to refine wording, she adds, and “a user may feel that attribution encroaches on their ownership if their AI use was limited.”

As companies and researchers attempt to better assign credit, or blame, for machine-aided work, people are grappling with the very human assumptions that their bosses and coworkers are making now. Multiple studies show that AI disclosure, even in good faith, can erode trust among colleagues and lead them to perceive those who use it as lazy.

Oliver Schilke, a management and sociology professor at the University of Arizona, whose research found that the simple act of disclosure can make people trust you less, agrees that the tension between firms’ urge to adopt AI for efficiency gains and the social costs associated with its adoption is one of the central contradictions of this new era of work. So far, Schilke says the burden has fallen on individual users, who must decide both to what extent to involve AI and to what extent to reveal this involvement, “creating the paradoxical dynamic that those who do the morally right thing must bear the penalty for transparency.” A better alternative, he adds, is collective AI governance norms that include tools such as the Attribution Toolkit.

Thomas Prommer, an engineering executive at Adidas, saw a similar pattern in his team. While mandatory AI attribution sounded fair, it quietly killed initiative for engineers. They quit reaching for AI tools, “because they didn’t want their best contributions footnoted as ‘cowritten by Claude,'” he says.

“The signal it sent was: AI help diminishes your work. So people hid it or avoided it,” says Prommer. What worked was crediting outcomes rather than tools. Irrespective of how much of the work was done by AI, ultimately, the person responsible gets the credit and the blame.

Many, including Aubrey, are starting to wonder if AI is just another program, like Excel, or if they need to credit it at all. Schilke disagrees with this mentality. While a traditional tool like Excel is designed to execute human instructions within a limited, predictable range, AI can produce prose, code, and ideas, often with very little human input. Beyond merely assisting in execution, Schilke adds, “it can make substantive contributions to the form and content of a work product, and that changes what it means to disclose its use.”

“The issue becomes whether the apparent intellectual contribution can still be attributed to the human author or agency was largely outsourced to the machine,” Schilke tells me.

For many workers, a bigger worry will be taking the fall for AI’s mistakes. Earlier this year, Amazon was found to have blamed humans and laid them off for an AI agent’s mistakes.

“The praise goes to AI, but going through its content is our responsibility, and if an error goes through, that, too, is registered as our responsibility,” adds Deepak.

Alessio Artuffo, the CEO of Docebo, a learning platform, says simple attribution is the wrong frame. The question is no longer how exactly the work was produced, but whether the person responsible for it can defend it, improve it, and be held accountable when it fails.

Ultimately, researchers and organizational experts agree that if companies want their employees to use AI in creative and productive ways, they will need to build environments where AI proficiency is something valuable to develop rather than something that puts them at risk of being passed over or blamed.

“The deeper cost is psychological,” adds Artuffo. “If employees are producing more output but feeling less ownership of the work, that’s not a win; that’s capability regression dressed up as efficiency.”